A Field Guide to

Genetic Programming
Riccardo Poli
Department of Computing and Electronic Systems
University of Essex UK
[email protected]
William B. Langdon
Departments of Biological and Mathematical Sciences
University of Essex UK
[email protected]
Nicholas F. McPhee
Division of Science and Mathematics
University of Minnesota, Morris USA
[email protected]
with contributions by
John R. Koza
Stanford University USA
[email protected]
March 2008

c
Riccardo
Poli, William B. Langdon, and Nicholas F. McPhee, 2008
This work is licensed under the Creative Commons AttributionNoncommercial-No Derivative Works 2.0 UK: England & Wales License
(see http://creativecommons.org/licenses/by-nc-nd/2.0/uk/). That
is:
You are free:
to copy, distribute, display, and perform the work
Under the following conditions:
Attribution. You must give the original authors credit.
Non-Commercial. You may not use this work for commercial
purposes.
No Derivative Works. You may not alter, transform, or build
upon this work.
For any reuse or distribution, you must make clear to others the licence
terms of this work. Any of these conditions can be waived if you get
permission from the copyright holders. Nothing in this license impairs
or restricts the authors rights.
Non-commercial uses are thus permitted without any further authorisation
from the copyright owners. The book may be freely downloaded in electronic
form at http://www.gp-field-guide.org.uk. Printed copies can also
be purchased inexpensively from http://lulu.com. For more information
about Creative Commons licenses, go to http://creativecommons.org
or send a letter to Creative Commons, 171 Second Street, Suite 300, San
Francisco, California, 94105, USA.

To cite this book, please see the entry for (Poli, Langdon, and McPhee,
2008) in the bibliography.

ISBN 978-1-4092-0073-4 (softcover)

Preface
Genetic programming (GP) is a collection of evolutionary computation techniques that allow computers to solve problems automatically. Since its inception twenty years ago, GP has been used to solve a wide range of practical problems, producing a number of human-competitive results and even
patentable new inventions. Like many other areas of computer science, GP
is evolving rapidly, with new ideas, techniques and applications being constantly proposed. While this shows how wonderfully prolific GP is, it also
makes it difficult for newcomers to become acquainted with the main ideas
in the field, and form a mental map of its different branches. Even for people
who have been interested in GP for a while, it is difficult to keep up with
the pace of new developments.
Many books have been written which describe aspects of GP. Some
provide general introductions to the field as a whole. However, no new
introductory book on GP has been produced in the last decade, and anyone
wanting to learn about GP is forced to map the terrain painfully on their
own. This book attempts to fill that gap, by providing a modern field guide
to GP for both newcomers and old-timers.
It would have been straightforward to find a traditional publisher for such
a book. However, we want our book to be as accessible as possible to everyone interested in learning about GP. Therefore, we have chosen to make it
freely available on-line, while also allowing printed copies to be ordered inexpensively from http://lulu.com. Visit http://www.gp-field-guide.
org.uk for the details.
The book has undergone numerous iterations and revisions. It began as
a book-chapter overview of GP (more on this below), which quickly grew
to almost 100 pages. A technical report version of it was circulated on the
GP mailing list. People responded very positively, and some encouraged us
to continue and expand that survey into a book. We took their advice and
this field guide is the result.

Acknowledgements
We would like to thank the University of Essex and the University of Minnesota, Morris, for their support.

Many thanks to Tyler Hutchison for the use of his cool drawing on the
cover (and elsewhere!), and for finding those scary pinks and greens.
We had the invaluable assistance of many people, and we are very grateful
for their individual and collective efforts, often on very short timelines. Rick
Riolo, Matthew Walker, Christian Gagne, Bob McKay, Giovanni Pazienza,
and Lee Spector all provided useful suggestions based on an early technical report version. Yossi Borenstein, Caterina Cinel, Ellery Crane, Cecilia
Di Chio, Stephen Dignum, Edgar Galv
an-L
opez, Keisha Harriott, David
Hunter, Lonny Johnson, Ahmed Kattan, Robert Keller, Andy Korth, Yevgeniya Kovalchuk, Simon Lucas, Wayne Manselle, Alberto Moraglio, Oliver
Oechsle, Francisco Sepulveda, Elias Tawil, Edward Tsang, William Tozier
and Christian Wagner all contributed to the final proofreading festival.
Their sharp eyes and hard work did much to make the book better; any
remaining errors or omissions are obviously the sole responsibility of the
authors.
We would also like to thank Prof. Xin Yao and the School of Computer
Science of The University of Birmingham and Prof. Bernard Buxton of University College, London, for continuing support, particularly of the genetic
programming bibliography. We also thank Schloss Dagstuhl, where some of
the integration of this book took place.
Most of the tools used in the construction of this book are open source,1
and we are very grateful to all the developers whose efforts have gone into
building those tools over the years.
As mentioned above, this book started life as a chapter. This was
for a forthcoming handbook on computational intelligence2 edited by John
Fulcher and Lakhmi C. Jain. We are grateful to John Fulcher for his useful
comments and edits on that book chapter. We would also like to thank most
warmly John Koza, who co-authored the aforementioned chapter with us,
and for allowing us to reuse some of his original material in this book.
This book is a summary of nearly two decades of intensive research in
the field of genetic programming, and we obviously owe a great debt to all
the researchers whose hard work, ideas, and interactions ultimately made
this book possible. Their work runs through every page, from an idea made
somewhat clearer by a conversation at a conference, to a specific concept
or diagram. It has been a pleasure to be part of the GP community over
the years, and we greatly appreciate having so much interesting work to
summarise!
March 2008

1 See

Riccardo Poli
William B. Langdon
Nicholas Freitag McPhee

the colophon (page 235) for more details.

entitled Computational Intelligence: A Compendium and to be published by Springer in 2008.
2 Tentatively

Whats in this book

The book is divided up into four parts.
Part I covers the basics of genetic programming (GP). This starts with a
gentle introduction which describes how a population of programs is stored
in the computer so that they can evolve with time. We explain how programs
are represented, how random programs are initially created, and how GP
creates a new generation by mutating the better existing programs or combining pairs of good parent programs to produce offspring programs. This
is followed by a simple explanation of how to apply GP and an illustrative
example of using GP.
In Part II, we describe a variety of alternative representations for programs and some advanced GP techniques. These include: the evolution of
machine-code and parallel programs, the use of grammars and probability
distributions for the generation of programs, variants of GP which allow the
solution of problems with multiple objectives, many speed-up techniques
and some useful theoretical tools.
Part III provides valuable information for anyone interested in using GP
in practical applications. To illustrate genetic programmings scope, this
part contains a review of many real-world applications of GP. These include: curve fitting, data modelling, symbolic regression, image analysis,
signal processing, financial trading, time series prediction, economic modelling, industrial process control, medicine, biology, bioinformatics, hyperheuristics, artistic applications, computer games, entertainment, compression and human-competitive results. This is followed by a series of recommendations and suggestions to obtain the most from a GP system. We then
provide some conclusions.
Part IV completes the book. In addition to a bibliography and an index,
this part includes two appendices that provide many pointers to resources,
further reading and a simple GP implementation in Java.

About the authors

The authors are experts in genetic programming with long and distinguished
track records, and over 50 years of combined experience in both theory and
practice in GP, with collaborations extending over a decade.
Riccardo Poli is a Professor in the Department of Computing and Electronic Systems at Essex. He started his academic career as an electronic engineer doing a PhD in biomedical image analysis to later become an expert
in the field of EC. He has published around 240 refereed papers and a book
(Langdon and Poli, 2002) on the theory and applications of genetic programming, evolutionary algorithms, particle swarm optimisation, biomedical engineering, brain-computer interfaces, neural networks, image/signal
processing, biology and psychology. He is a Fellow of the International Society for Genetic and Evolutionary Computation (2003), a recipient of the
EvoStar award for outstanding contributions to this field (2007), and an
ACM SIGEVO executive board member (20072013). He was co-founder
and co-chair of the European Conference on GP (19982000, 2003). He was
general chair (2004), track chair (2002, 2007), business committee member
(2005), and competition chair (2006) of ACMs Genetic and Evolutionary
Computation Conference, co-chair of the Foundations of Genetic Algorithms
Workshop (2002) and technical chair of the International Workshop on Ant
Colony Optimisation and Swarm Intelligence (2006). He is an associate editor of Genetic Programming and Evolvable Machines, Evolutionary Computation and the International Journal of Computational Intelligence Research.
He is an advisory board member of the Journal on Artificial Evolution and
Applications and an editorial board member of Swarm Intelligence. He is a
member of the EPSRC Peer Review College, an EU expert evaluator and a
grant-proposal referee for Irish, Swiss and Italian funding bodies.
W. B. Langdon was research officer for the Central Electricity Research
Laboratories and project manager and technical coordinator for Logica before becoming a prolific, internationally recognised researcher (working at
UCL, Birmingham, CWI and Essex). He has written two books, edited
six more, and published over 80 papers in international conferences and
journals. He is the resource review editor for Genetic Programming and
Evolvable Machines and a member of the editorial board of Evolutionary

Computation. He has been a co-organiser of eight international conferences

and workshops, and has given nine tutorials at international conferences. He
was elected ISGEC Fellow for his contributions to EC. Dr Langdon has extensive experience designing and implementing GP systems, and is a leader
in both the empirical and theoretical analysis of evolutionary systems. He
also has broad experience both in industry and academic settings in biomedical engineering, drug design, and bioinformatics.
Nicholas F. McPhee is a Full Professor in Computer Science in the
Division of Science and Mathematics, University of Minnesota, Morris. He
is an associate editor of the Journal on Artificial Evolution and Applications, an editorial board member of Genetic Programming and Evolvable
Machines, and has served on the program committees for dozens of international events. He has extensive expertise in the design of GP systems, and in
the theoretical analysis of their behaviours. His joint work with Poli on the
theoretical analysis of GP (McPhee and Poli, 2001; Poli and McPhee, 2001)
received the best paper award at the 2001 European Conference on Genetic
Programming, and several of his other foundational studies continue to be
widely cited. He has also worked closely with biologists on a number of
projects, building individual-based models to illuminate genetic interactions
and changes in the genotypic and phenotypic diversity of populations.

To
Caterina, Ludovico, Rachele and Leonardo
Susan and Thomas

R.P.

N.F.M.

Contents
Contents

xi

1 Introduction
1.1 Genetic Programming in a Nutshell .
1.2 Getting Started . . . . . . . . . . . .
1.3 Prerequisites . . . . . . . . . . . . .
1.4 Overview of this Field Guide . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

Basics

1
2
2
3
4

2 Representation, Initialisation and

GP
2.1 Representation . . . . . . . . . .
2.2 Initialising the Population . . . .
2.3 Selection . . . . . . . . . . . . . .
2.4 Recombination and Mutation . .

Operators in Tree-based
.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

9
9
11
14
15

3 Getting Ready to Run Genetic Programming

3.1 Step 1: Terminal Set . . . . . . . . . . . . . . . .
3.2 Step 2: Function Set . . . . . . . . . . . . . . . .
3.2.1 Closure . . . . . . . . . . . . . . . . . . .
3.2.2 Sufficiency . . . . . . . . . . . . . . . . . .
3.2.3 Evolving Structures other than Programs
3.3 Step 3: Fitness Function . . . . . . . . . . . . . .
3.4 Step 4: GP Parameters . . . . . . . . . . . . . .
3.5 Step 5: Termination and solution designation . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

19
19
20
21
23
23
24
26
27

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

4 Example Genetic Programming Run

29
4.1 Preparatory Steps . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2 Step-by-Step Sample Run . . . . . . . . . . . . . . . . . . . . 31
4.2.1 Initialisation . . . . . . . . . . . . . . . . . . . . . . . 31
xi

CONTENTS
4.2.2
4.2.3
4.2.4

II

CONTENTS
Fitness Evaluation . . . . . . . . . . . . . . . . . . . .
Selection, Crossover and Mutation . . . . . . . . . . .
Termination and Solution Designation . . . . . . . . .

Advanced Genetic Programming

5 Alternative Initialisations and Operators in

5.1 Constructing the Initial Population . . . . .
5.1.1 Uniform Initialisation . . . . . . . .
5.1.2 Initialisation may Affect Bloat . . .
5.1.3 Seeding . . . . . . . . . . . . . . . .
5.2 GP Mutation . . . . . . . . . . . . . . . . .
5.2.1 Is Mutation Necessary? . . . . . . .
5.2.2 Mutation Cookbook . . . . . . . . .
5.3 GP Crossover . . . . . . . . . . . . . . . . .
5.4 Other Techniques . . . . . . . . . . . . . . .

32
32
35

37
Tree-based GP
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .
. . . . . . . . . .

39
39
40
40
41
42
42
42
44
46

6 Modular, Grammatical and Developmental Tree-based GP

6.1 Evolving Modular and Hierarchical Structures . . . . . . . . .
6.1.1 Automatically Defined Functions . . . . . . . . . . . .
6.1.2 Program Architecture and Architecture-Altering . . .
6.2 Constraining Structures . . . . . . . . . . . . . . . . . . . . .
6.2.1 Enforcing Particular Structures . . . . . . . . . . . . .
6.2.2 Strongly Typed GP . . . . . . . . . . . . . . . . . . .
6.2.3 Grammar-based Constraints . . . . . . . . . . . . . . .
6.2.4 Constraints and Bias . . . . . . . . . . . . . . . . . . .
6.3 Developmental Genetic Programming . . . . . . . . . . . . .
6.4 Strongly Typed Autoconstructive GP with PushGP . . . . .

47
47
48
50
51
52
52
53
55
57
59

7 Linear and Graph Genetic Programming

7.1 Linear Genetic Programming . . . . . . . . . . . . . . . . .
7.1.1 Motivations . . . . . . . . . . . . . . . . . . . . . . .
7.1.2 Linear GP Representations . . . . . . . . . . . . . .
7.1.3 Linear GP Operators . . . . . . . . . . . . . . . . . .
7.2 Graph-Based Genetic Programming . . . . . . . . . . . . .
7.2.1 Parallel Distributed GP (PDGP) . . . . . . . . . . .
7.2.2 PADO . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2.3 Cartesian GP . . . . . . . . . . . . . . . . . . . . . .
7.2.4 Evolving Parallel Programs using Indirect Encodings

61
61
61
62
64
65
65
67
67
68

xii

.
.
.
.
.
.
.
.
.

CONTENTS

CONTENTS

8 Probabilistic Genetic Programming

69
8.1 Estimation of Distribution Algorithms . . . . . . . . . . . . . 69
8.2 Pure EDA GP . . . . . . . . . . . . . . . . . . . . . . . . . . 71
8.3 Mixing Grammars and Probabilities . . . . . . . . . . . . . . 74
9 Multi-objective Genetic Programming
9.1 Combining Multiple Objectives into a Scalar Fitness Function
9.2 Keeping the Objectives Separate . . . . . . . . . . . . . . . .
9.2.1 Multi-objective Bloat and Complexity Control . . . .
9.2.2 Other Objectives . . . . . . . . . . . . . . . . . . . . .
9.2.3 Non-Pareto Criteria . . . . . . . . . . . . . . . . . . .
9.3 Multiple Objectives via Dynamic and Staged Fitness Functions
9.4 Multi-objective Optimisation via Operator Bias . . . . . . . .

75
75
76
77
78
80
80
81

10 Fast and Distributed Genetic Programming

10.1 Reducing Fitness Evaluations/Increasing their Effectiveness
10.2 Reducing Cost of Fitness with Caches . . . . . . . . . . . .
10.3 Parallel and Distributed GP are Not Equivalent . . . . . . .
10.4 Running GP on Parallel Hardware . . . . . . . . . . . . . .
10.4.1 Masterslave GP . . . . . . . . . . . . . . . . . . . .
10.4.2 GP Running on GPUs . . . . . . . . . . . . . . . . .
10.4.3 GP on FPGAs . . . . . . . . . . . . . . . . . . . . .
10.4.4 Sub-machine-code GP . . . . . . . . . . . . . . . . .
10.5 Geographically Distributed GP . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.

83
83
86
88
89
89
90
92
93
93

11 GP Theory and its Applications

11.1 Mathematical Models . . . . . .
11.2 Search Spaces . . . . . . . . . . .
11.3 Bloat . . . . . . . . . . . . . . . .
11.3.1 Bloat in Theory . . . . .
11.3.2 Bloat Control in Practice

97
. 98
. 99
. 101
. 101
. 104

III

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

Practical Genetic Programming

12 Applications
12.1 Where GP has Done Well . . . . . . . . . . . . . . . . . .
12.2 Curve Fitting, Data Modelling and Symbolic Regression .
12.3 Human Competitive Results the Humies . . . . . . . . .
12.4 Image and Signal Processing . . . . . . . . . . . . . . . . .
12.5 Financial Trading, Time Series, and Economic Modelling
12.6 Industrial Process Control . . . . . . . . . . . . . . . . . .
12.7 Medicine, Biology and Bioinformatics . . . . . . . . . . .
12.8 GP to Create Searchers and Solvers Hyper-heuristics . .
xiii

109
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

111
111
113
117
121
123
124
125
126

CONTENTS

CONTENTS

12.9 Entertainment and Computer Games . . . . . . . . . . . . . . 127

12.10The Arts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
12.11Compression . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
13 Troubleshooting GP
13.1 Is there a Bug in the Code? . . . . .
13.2 Can you Trust your Results? . . . .
13.3 There are No Silver Bullets . . . . .
13.4 Small Changes can have Big Effects
13.5 Big Changes can have No Effect . .
13.6 Study your Populations . . . . . . .
13.7 Encourage Diversity . . . . . . . . .
13.8 Embrace Approximation . . . . . . .
13.9 Control Bloat . . . . . . . . . . . . .
13.10Checkpoint Results . . . . . . . . . .
13.11Report Well . . . . . . . . . . . . . .
13.12Convince your Customers . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

14 Conclusions

IV

131
131
132
132
133
133
134
136
137
139
139
139
140
141

Tricks of the Trade

143

A Resources
A.1 Key Books . . . . . . . . . .
A.2 Key Journals . . . . . . . .
A.3 Key International Meetings
A.4 GP Implementations . . . .
A.5 On-Line Resources . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

145
146
147
147
147
148

B TinyGP
B.1 Overview of TinyGP . . . . . . .
B.2 Input Data Files for TinyGP . .
B.3 Source Code . . . . . . . . . . . .
B.4 Compiling and Running TinyGP

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

151
151
153
154
162

.
.
.
.
.

.
.
.
.
.

Bibliography

167

Index

225

xiv

Chapter 1

Introduction
The goal of having computers automatically solve problems is central to
artificial intelligence, machine learning, and the broad area encompassed by
what Turing called machine intelligence (Turing, 1948). Machine learning
pioneer Arthur Samuel, in his 1983 talk entitled AI: Where It Has Been
and Where It Is Going (Samuel, 1983), stated that the main goal of the
fields of machine learning and artificial intelligence is:
to get machines to exhibit behaviour, which if done by humans,
would be assumed to involve the use of intelligence.
Genetic programming (GP) is an evolutionary computation (EC)1 technique that automatically solves problems without requiring the user to know
or specify the form or structure of the solution in advance. At the most
abstract level GP is a systematic, domain-independent method for getting
computers to solve problems automatically starting from a high-level statement of what needs to be done.
Since its inception, GP has attracted the interest of myriads of people
around the globe. This book gives an overview of the basics of GP, summarised important work that gave direction and impetus to the field and
discusses some interesting new directions and applications. Things continue
to change rapidly in genetic programming as investigators and practitioners
discover new methods and applications. This makes it impossible to cover
all aspects of GP, and this book should be seen as a snapshot of a particular
moment in the history of the field.

1 These

are also known as evolutionary algorithms or EAs.

1 Introduction

Generate Population
of Random Programs

Run Programs and

Evaluate Their Quality

Solution
(* (SIN (- y x))
(IF (> x 15.43)
(+ 2.3787 x)
(* (SQRT y)
(/ x 7.54))))

Breed Fitter Programs

Figure 1.1: The basic control flow for genetic programming, where survival
of the fittest is used to find solutions.

1.1

Genetic Programming in a Nutshell

In genetic programming we evolve a population of computer programs. That

is, generation by generation, GP stochastically transforms populations of
programs into new, hopefully better, populations of programs, cf. Figure 1.1.
GP, like nature, is a random process, and it can never guarantee results.
GPs essential randomness, however, can lead it to escape traps which deterministic methods may be captured by. Like nature, GP has been very
successful at evolving novel and unexpected ways of solving problems. (See
Chapter 12 for numerous examples.)
The basic steps in a GP system are shown in Algorithm 1.1. GP finds out
how well a program works by running it, and then comparing its behaviour
to some ideal (line 3). We might be interested, for example, in how well a
program predicts a time series or controls an industrial process. This comparison is quantified to give a numeric value called fitness. Those programs
that do well are chosen to breed (line 4) and produce new programs for the
next generation (line 5). The primary genetic operations that are used to
create new programs from existing ones are:
Crossover: The creation of a child program by combining randomly
chosen parts from two selected parent programs.
Mutation: The creation of a new child program by randomly altering
a randomly chosen part of a selected parent program.

1.2

Getting Started

Two key questions for those first exploring GP are:

1. What should I read to get started in GP?
2. Should I implement my own GP system or should I use an existing
package? If so, what package should I use?

1.3 Prerequisites
1:
2:
3:
4:
5:
6:
7:

Randomly create an initial population of programs from the available

primitives (more on this in Section 2.2).
repeat
Execute each program and ascertain its fitness.
Select one or two program(s) from the population with a probability
based on fitness to participate in genetic operations (Section 2.3).
Create new individual program(s) by applying genetic operations with
specified probabilities (Section 2.4).
until an acceptable solution is found or some other stopping condition
is met (e.g., a maximum number of generations is reached).
return the best-so-far individual.
Algorithm 1.1: Genetic Programming

The best way to begin is obviously by reading this book, so youre off to
a good start. We included a wide variety of references to help guide people
through at least some of the literature. No single work, however, could claim
to be completely comprehensive. Thus Appendix A reviews a whole host of
books, videos, journals, conferences, and on-line sources (including several
freely available GP systems) that should be of assistance.
We strongly encourage doing GP as well as reading about it; the dynamics of evolutionary algorithms are complex, and the experience of tracing through runs is invaluable. In Appendix B we provide the full Java
implementation of Riccardos TinyGP system.

1.3

Prerequisites

Although this book has been written with beginners in mind, unavoidably
we had to make some assumptions about the typical background of our
readers. The book assumes some working knowledge of computer science
and computer programming; this is probably an essential prerequisite to get
the most from the book.
We dont expect that readers will have been exposed to other flavours of
evolutionary algorithms before, although a little background might be useful.
The interested novice can easily find additional information on evolutionary
computation thanks to the plethora of tutorials available on the Internet.
Articles from Wikipedia and the genetic algorithm tutorial produced by
Whitley (1994) should suffice.

1.4

1 Introduction

Overview of this Field Guide

As we indicated in the section entitled Whats in this book (page v), the
book is divided up into four parts. In this section, we will have a closer look
at their content.
Part I is mainly for the benefit of beginners, so notions are introduced
at a relaxed pace. In the next chapter we provide a description of the key
elements in GP. These include how programs are stored (Section 2.1), the
initialisation of the population (Section 2.2), the selection of individuals
(Section 2.3) and the genetic operations of crossover and mutation (Section 2.4). A discussion of the decisions that are needed before running GP
is given in Chapter 3. These preparatory steps include the specification of
the set of instructions that GP can use to construct programs (Sections 3.1
and 3.2), the definition of a fitness measure that can guide GP towards
good solutions (Section 3.3), setting GP parameters (Section 3.4) and, finally, the rule used to decide when to stop a GP run (Section 3.5). To help
the reader understand these, Chapter 4 presents a step-by-step application
of the preparatory steps (Section 4.1) and a detailed explanation of a sample
GP run (Section 4.2).
After these introductory chapters, we go up a gear in Part II where
we describe a variety of more advanced GP techniques. Chapter 5 considers additional initialisation strategies and genetic operators for the main GP
representationsyntax trees. In Chapter 6 we look at techniques for the evolution of structured and grammatically-constrained programs. In particular,
we consider: modular and hierarchical structures including automatically defined functions and architecture-altering operations (Section 6.1), systems
that constrain the syntax of evolved programs using grammars or type systems (Section 6.2), and developmental GP (Section 6.3). In Chapter 7 we
discuss alternative program representations, namely linear GP (Section 7.1)
and graph-based GP (Section 7.2).
In Chapter 8 we review systems where, instead of using mutation and
recombination to create new programs, they are simply generated randomly
according to a probability distribution which itself evolves. These are known
as estimation of distribution algorithms, cf. Sections 8.1 and 8.2. Section 8.3
reviews hybrids between GP and probabilistic grammars, where probability
distributions are associated with the elements of a grammar.
Many, if not most, real-world problems are multi-objective, in the sense
that their solutions are required to satisfy more than one criterion at the
same time. In Chapter 9, we review different techniques that allow GP to
solve multi-objective problems. These include the aggregation of multiple
objectives into a scalar fitness measure (Section 9.1), the use of the notion of
Pareto dominance (Section 9.2), the definition of dynamic or staged fitness
functions (Section 9.3), and the reliance on special biases on the genetic
operators to aid the optimisation of multiple objectives (Section 9.4).

1.4 Overview of this Field Guide

A variety of methods to speed up, parallelise and distribute genetic programming runs are described in Chapter 10. We start by looking at ways
to reduce the number of fitness evaluations or increase their effectiveness
(Section 10.1) and ways to speed up their execution (Section 10.2). We
then point out (Section 10.3) that faster evaluation is not the only reason
for running GP in parallel, as geographic distribution has advantages in
its own right. In Section 10.4, we consider the first approach and describe
master-slave parallel architectures (Section 10.4.1), running GP on graphics
hardware (Section 10.4.2) and FPGAs (Section 10.4.3), and a fast method to
exploit the parallelism available on every computer (Section 10.4.4). Finally,
Section 10.5 looks at the second approach discussing the geographically distributed evolution of programs. We then give an overview of some of the
considerable work that has been done on GPs theory and its practical uses
(Chapter 11).
After this review of techniques, Part III provides information for people interested in using GP in practical applications. We survey the enormous variety of applications of GP in Chapter 12. We start with a discussion of the general kinds of problems where GP has proved successful
(Section 12.1) and then describe a variety of GP applications, including:
curve fitting, data modelling and symbolic regression (Section 12.2); human
competitive results (Section 12.3); image analysis and signal processing (Section 12.4); financial trading, time series prediction and economic modelling
(Section 12.5); industrial process control (Section 12.6); medicine, biology
and bioinformatics (Section 12.7); the evolution of search algorithms and
optimisers (Section 12.8); computer games and entertainment applications
(Section 12.9); artistic applications (12.10); and GP-based data compression
(Section 12.11). This is followed by a chapter providing a collection of troubleshooting techniques used by experienced GP practitioners (Chapter 13)
and by our conclusions (Chapter 14).
In Part IV, we provide a resources appendix that reviews the many
sources of further information on GP, on its applications, and on related
problem solving systems (Appendix A). This is followed by a description
and the source code for a simple GP system in Java (Appendix B). The
results of a sample run with the system are also described in the appendix
and further illustrated via a Flip-O-Rama animation2 (see Section B.4).
The book ends with a large bibliography containing around 650 references. Of these, around 420 contain pointers to on-line versions of the corresponding papers. While this is very useful on its own, the users of the PDF
version of this book will be able to do more if they use a PDF viewer that
supports hyperlinks: they will be able to click on the URLs and retrieve the
cited articles. Around 550 of the papers in the bibliography are included in

2 This

is in the footer of the odd-numbered pages in the bibliography and in the index.

1 Introduction

the GP bibliography (Langdon, Gustafson, and Koza, 1995-2008).3 We have

linked those references to the corresponding BibTEXentries in the bibliography. Just click on the GPBiB symbols to retrieve them instantaneously.
Entries in the bibliography typically include keywords, abstracts and often
further URLs.
With a slight self-referential violation of bibliographic etiquette, we have
also included in the bibliography the excellent (Poli et al., 2008) to clarify how to cite this book. LATEX users can find the BibTEX entry for
this book at http://www.cs.bham.ac.uk/~wbl/biblio/gp-html/poli08_
fieldguide.html.

3 Available

at http://www.cs.bham.ac.uk/~wbl/biblio/

Part I

Basics

Here Alice steps through the looking glass. . .

and the Jabberwock is slain.

Chapter 2

Representation,
Initialisation and
Operators in Tree-based
GP
This chapter introduces the basic tools and terminology used in genetic
programming. In particular, it looks at how trial solutions are represented in
most GP systems (Section 2.1), how one might construct the initial random
population (Section 2.2), and how selection (Section 2.3) as well as crossover
and mutation (Section 2.4) are used to construct new programs.

2.1

Representation

In GP, programs are usually expressed as syntax trees rather than as lines of
code. For example Figure 2.1 shows the tree representation of the program
max(x+x,x+3*y). The variables and constants in the program (x, y and 3)
are leaves of the tree. In GP they are called terminals, whilst the arithmetic
operations (+, * and max) are internal nodes called functions. The sets of
allowed functions and terminals together form the primitive set of a GP
system.
In more advanced forms of GP, programs can be composed of multiple
components (e.g., subroutines). In this case the representation used in GP
is a set of trees (one for each component) grouped together under a special
root node that acts as glue, as illustrated in Figure 2.2. We will call these
(sub)trees branches. The number and type of the branches in a program,
9

10

2 Tree-based GP

together with certain other features of their structure, form the architecture
of the program. This is discussed in more detail in Section 6.1.
It is common in the GP literature to represent expressions in a prefix notation similar to that used in Lisp or Scheme. For example, max(x+x,x+3*y)
becomes (max (+ x x) (+ x (* 3 y))). This notation often makes it easier to see the relationship between (sub)expressions and their corresponding
(sub)trees. Therefore, in the following, we will use trees and their corresponding prefix-notation expressions interchangeably.
How one implements GP trees will obviously depend a great deal on
the programming languages and libraries being used. Languages that provide automatic garbage collection and dynamic lists as fundamental data
types make it easier to implement expression trees and the necessary GP
operations. Most traditional languages used in AI research (e.g., Lisp and
Prolog), many recent languages (e.g., Ruby and Python), and the languages
associated with several scientific programming tools (e.g., MATLAB1 and
Mathematica2 ) have these facilities. In other languages, one may have to
implement lists/trees or use libraries that provide such data structures.
In high performance environments, the tree-based representation of programs may be too inefficient since it requires the storage and management
of numerous pointers. In some cases, it may be desirable to use GP primitives which accept a variable number of arguments (a quantity we will call
arity). An example is the sequencing instruction progn, which accepts any
number of arguments, executes them one at a time and then returns the

max
+
x

+
x

x
3

Figure 2.1: GP syntax tree representing max(x+x,x+3*y).

1 MATLAB

is a registered trademark of The MathWorks, Inc

is a registered trademark of Wolfram Research, Inc.

2 Mathematica

2.2 Initialising the Population

11

ROOT
Branches

...
Component
1

Component
2

Component
N

Figure 2.2: Multi-component program representation.

value returned by the last argument. However, fortunately, it is now extremely common in GP applications for all functions to have a fixed number
of arguments. If this is the case, then, the brackets in prefix-notation expressions are redundant, and trees can efficiently be represented as simple
linear sequences. In effect, the functions name gives its arity and from the
arities the brackets can be inferred. For example, the expression (max (+ x
x) (+ x (* 3 y))) could be written unambiguously as the sequence max
+ x x + x * 3 y.
The choice of whether to use such a linear representation or an explicit
tree representation is typically guided by questions of convenience, efficiency,
the genetic operations being used (some may be more easily or more efficiently implemented in one representation), and other data one may wish
to collect during runs. (It is sometimes useful to attach additional information to nodes, which may be easier to implement if they are explicitly
represented).
These tree representations are the most common in GP, e.g., numerous high-quality, freely available GP implementations use them (see the
resources in Appendix A, page 148, for more information) and so does also
the simple GP system described in Appendix B. However, there are other
important representations, some of which are discussed in Chapter 7.

2.2

Initialising the Population

Like in other evolutionary algorithms, in GP the individuals in the initial

population are typically randomly generated. There are a number of different approaches to generating this random initial population. Here we

12

2 Tree-based GP
t=1

t=2
+

t=3

t=4

x
t=5

t=6

t=7

y 1

Figure 2.3: Creation of a full tree having maximum depth 2 using the full
initialisation method (t = time).

will describe two of the simplest (and earliest) methods (the full and grow
methods), and a widely used combination of the two known as Ramped halfand-half.
In both the full and grow methods, the initial individuals are generated
so that they do not exceed a user specified maximum depth. The depth of
a node is the number of edges that need to be traversed to reach the node
starting from the trees root node (which is assumed to be at depth 0). The
depth of a tree is the depth of its deepest leaf (e.g., the tree in Figure 2.1 has
a depth of 3). In the full method (so named because it generates full trees,
i.e. all leaves are at the same depth) nodes are taken at random from the
function set until the maximum tree depth is reached. (Beyond that depth,
only terminals can be chosen.) Figure 2.3 shows a series of snapshots of the
construction of a full tree of depth 2. The children of the * and / nodes
must be leaves or otherwise the tree would be too deep. Thus, at both steps
t = 3, t = 4, t = 6 and t = 7 a terminal must be chosen (x, y, 1 and 0,
respectively).
Although, the full method generates trees where all the leaves are at
the same depth, this does not necessarily mean that all initial trees will
have an identical number of nodes (often referred to as the size of a tree)
or the same shape. This only happens, in fact, when all the functions in
the primitive set have an equal arity. Nonetheless, even when mixed-arity
primitive sets are used, the range of program sizes and shapes produced by
the full method may be rather limited. The grow method, on the contrary,
allows for the creation of trees of more varied sizes and shapes. Nodes are
selected from the whole primitive set (i.e., functions and terminals) until
the depth limit is reached. Once the depth limit is reached only terminals

2.2 Initialising the Population

13

t=1

t=2
+

t=3
+

x
t=4

t=5

x
2

x
2

Figure 2.4: Creation of a five node tree using the grow initialisation method
with a maximum depth of 2 (t = time). A terminal is chosen at t = 2,
causing the left branch of the root to be closed at that point even though
the maximum depth had not been reached.

may be chosen (just as in the full method). Figure 2.4 illustrates this
process for the construction of a tree with depth limit 2. Here the first
argument of the + root node happens to be a terminal. This closes off that
branch preventing it from growing any more before it reached the depth
limit. The other argument is a function (-), but its arguments are forced
to be terminals to ensure that the resulting tree does not exceed the depth
limit. Pseudocode for a recursive implementation of both the full and grow
methods is given in Algorithm 2.1.
Because neither the grow or full method provide a very wide array of
sizes or shapes on their own, Koza (1992) proposed a combination called
ramped half-and-half. Half the initial population is constructed using full
and half is constructed using grow. This is done using a range of depth limits
(hence the term ramped) to help ensure that we generate trees having a
variety of sizes and shapes.
While these methods are easy to implement and use, they often make it
difficult to control the statistical distributions of important properties such
as the sizes and shapes of the generated trees. For example, the sizes and
shapes of the trees generated via the grow method are highly sensitive to the
sizes of the function and terminal sets. If, for example, one has significantly
more terminals than functions, the grow method will almost always generate
very short trees regardless of the depth limit. Similarly, if the number of
functions is considerably greater than the number of terminals, then the
grow method will behave quite similarly to the full method. The arities
of the functions in the primitive set also influence the size and shape of the

14

2 Tree-based GP

procedure: gen rnd expr(func set,term set,max d,method)



|term set|
1: if max d = 0 or method = grow and rand() <
|term set|+|func set|
then
2:
expr = choose random element( term set )
3: else
4:
func = choose random element( func set )
5:
for i = 1 to arity(func) do
6:
arg i = gen rnd expr( func set, term set, max d - 1, method );
7:
end for
8:
expr = (func, arg 1, arg 2, ...);
9: end if
10: return expr
Notes: func set is a function set, term set is a terminal set, max d is the
maximum allowed depth for expressions, method is either full or grow,
expr is the generated expression in prefix notation and rand() is a function
that returns random numbers uniformly distributed between 0 and 1.
Algorithm 2.1: Pseudocode for recursive program generation with the
full and grow methods.
trees produced by grow.3 Section 5.1 (page 40) describes other initialisation
mechanisms which address these issues.
The initial population need not be entirely random. If something is
known about likely properties of the desired solution, trees having these
properties can be used to seed the initial population. This, too, will be
described in Section 5.1.

2.3

Selection

As with most evolutionary algorithms, genetic operators in GP are applied

to individuals that are probabilistically selected based on fitness. That is,
better individuals are more likely to have more child programs than inferior
individuals. The most commonly employed method for selecting individuals
in GP is tournament selection, which is discussed below, followed by fitnessproportionate selection, but any standard evolutionary algorithm selection
mechanism can be used.
In tournament selection a number of individuals are chosen at random
3 While these are particular problems for the grow method, they illustrate a general
issue where small (and often apparently inconsequential) changes such as the addition or
removal of a few functions from the function set can in fact have significant implications
for the GP system, and potentially introduce important but unintended biases.

2.4 Recombination and Mutation

15

from the population. These are compared with each other and the best of
them is chosen to be the parent. When doing crossover, two parents are
needed and, so, two selection tournaments are made. Note that tournament selection only looks at which program is better than another. It does
not need to know how much better. This effectively automatically rescales
fitness, so that the selection pressure4 on the population remains constant.
Thus, a single extraordinarily good program cannot immediately swamp the
next generation with its children; if it did, this would lead to a rapid loss
of diversity with potentially disastrous consequences for a run. Conversely,
tournament selection amplifies small differences in fitness to prefer the better program even if it is only marginally superior to the other individuals in
a tournament.
An element of noise is inherent in tournament selection due to the random selection of candidates for tournaments. So, while preferring the best,
tournament selection does ensure that even average-quality programs have
some chance of having children. Since tournament selection is easy to implement and provides automatic fitness rescaling, it is commonly used in GP.
Considering that selection has been described many times in the evolutionary algorithms literature, we will not provide details of the numerous
other mechanisms that have been proposed. (Goldberg, 1989), for example,
describes fitness-proportionate selection, stochastic universal sampling and
several others.

2.4

Recombination and Mutation

GP departs significantly from other evolutionary algorithms in the implementation of the operators of crossover and mutation. The most commonly
used form of crossover is subtree crossover. Given two parents, subtree
crossover randomly (and independently) selects a crossover point (a node)
in each parent tree. Then, it creates the offspring by replacing the subtree
rooted at the crossover point in a copy of the first parent with a copy of
the subtree rooted at the crossover point in the second parent, as illustrated
in Figure 2.5. Copies are used to avoid disrupting the original individuals.
This way, if selected multiple times, they can take part in the creation of
multiple offspring programs. Note that it is also possible to define a version
of crossover that returns two offspring, but this is not commonly used.
Often crossover points are not selected with uniform probability. Typical
GP primitive sets lead to trees with an average branching factor (the number of children of each node) of at least two, so the majority of the nodes
will be leaves. Consequently the uniform selection of crossover points leads
4 A key property of any selection mechanism is selection pressure. A system with a
strong selection pressure very highly favours the more fit individuals, while a system with
a weak selection pressure isnt so discriminating.

16

2 Tree-based GP

Parents
Crossover
Point

(x+y)+3

Crossover
Point

+
y

Offspring

(x/2)+3
3

(y+1)* (x/2)
/

1x

GARBAGE

Figure 2.5: Example of subtree crossover. Note that the trees on the left
are actually copies of the parents. So, their genetic material can freely be
used without altering the original individuals.

to crossover operations frequently exchanging only very small amounts of

genetic material (i.e., small subtrees); many crossovers may in fact reduce
to simply swapping two leaves. To counter this, Koza (1992) suggested the
widely used approach of choosing functions 90% of the time and leaves 10%
of the time. Many other types of crossover and mutation of GP trees are
possible. They will be described in Sections 5.2 and 5.3, pages 4246.
The most commonly used form of mutation in GP (which we will call
subtree mutation) randomly selects a mutation point in a tree and substitutes the subtree rooted there with a randomly generated subtree. This is
illustrated in Figure 2.6. Subtree mutation is sometimes implemented as
crossover between a program and a newly generated random program; this
operation is also known as headless chicken crossover (Angeline, 1997).
Another common form of mutation is point mutation, which is GPs
rough equivalent of the bit-flip mutation used in genetic algorithms (Goldberg, 1989). In point mutation, a random node is selected and the primitive
stored there is replaced with a different random primitive of the same arity
taken from the primitive set. If no other primitives with that arity exist, nothing happens to that node (but other nodes may still be mutated).
When subtree mutation is applied, this involves the modification of exactly
one subtree. Point mutation, on the other hand, is typically applied on a

2.4 Recombination and Mutation

17

Parents
Mutation
Point

+
+
x

Offspring

3
y

Mutation
Point

Randomly Generated
Sub-tree

y
x

y
x

Figure 2.6: Example of subtree mutation.

per-node basis. That is, each node is considered in turn and, with a certain
probability, it is altered as explained above. This allows multiple nodes to
be mutated independently in one application of point mutation.
The choice of which of the operators described above should be used
to create an offspring is probabilistic. Operators in GP are normally mutually exclusive (unlike other evolutionary algorithms where offspring are
sometimes obtained via a composition of operators). Their probability of
application are called operator rates. Typically, crossover is applied with the
highest probability, the crossover rate often being 90% or higher. On the
contrary, the mutation rate is much smaller, typically being in the region of
1%.
When the rates of crossover and mutation add up to a value p which is
less than 100%, an operator called reproduction is also used, with a rate of
1 p. Reproduction simply involves the selection of an individual based on
fitness and the insertion of a copy of it in the next generation.

Chapter 3

Getting Ready to Run

Genetic Programming
To apply a GP system to a problem, several decisions need to be made;
these are often termed the preparatory steps. The key choices are:
1. What it the terminal set?
2. What is the function set?
3. What is the fitness measure?
4. What parameters will be used for controlling the run?
5. What will be the termination criterion, and what will be designated
the result of the run?

3.1

Step 1: Terminal Set

While it is common to describe GP as evolving programs, GP is not typically used to evolve programs in the familiar Turing-complete languages
humans normally use for software development. It is instead more common to evolve programs (or expressions or formulae) in a more constrained
and often domain-specific language. The first two preparatory steps, the
definition of the terminal and function sets, specify such a language. That
is, together they define the ingredients that are available to GP to create
computer programs.
19

20

3 Getting Ready to Run Genetic Programming

The terminal set may consist of:
the programs external inputs. These typically take the form of named
variables (e.g., x, y).
functions with no arguments. These may be included because they
return different values each time they are used, such as the function
rand() which returns random numbers, or a function dist to wall()
that returns the distance to an obstacle from a robot that GP is controlling. Another possible reason is because the function produces side
effects. Functions with side effects do more than just return a value:
they may change some global data structures, print or draw something
on the screen, control the motors of a robot, etc.
constants. These can be pre-specified, randomly generated as part of
the tree creation process, or created by mutation.

Using a primitive such as rand can cause the behaviour of an individual

program to vary every time it is called, even if it is given the same inputs.
This is desirable in some applications. However, we more often want a
set of fixed random constants that are generated as part of the process of
initialising the population. This is typically accomplished by introducing
a terminal that represents an ephemeral random constant. Every time this
terminal is chosen in the construction of an initial tree (or a new subtree
to use in an operation like mutation), a different random value is generated
which is then used for that particular terminal, and which will remain fixed
for the rest of the run. The use of ephemeral random constants is typically
denoted by including the symbol < in the terminal set; see Chapter 4 for an
example.

3.2

Step 2: Function Set

The function set used in GP is typically driven by the nature of the problem
domain. In a simple numeric problem, for example, the function set may
consist of merely the arithmetic functions (+, -, *, /). However, all sorts
of other functions and constructs typically encountered in computer programs can be used. Table 3.1 shows a sample of some of the functions one
sees in the GP literature. Sometimes the primitive set includes specialised
functions and terminals which are designed to solve problems in a specific
problem domain. For example, if the goal is to program a robot to mop the
floor, then the function set might include such actions as move, turn, and
swish-the-mop.

3.2 Step 2: Function Set

21

Table 3.1: Examples of primitives in GP function and terminal sets.

Function Set
Kind of Primitive Example(s)
Arithmetic
+, *, /
Mathematical
sin, cos, exp
Boolean
AND, OR, NOT
Conditional
IF-THEN-ELSE
Looping
FOR, REPEAT
..
..
.
.
Terminal Set
Kind of Primitive Example(s)
Variables
x, y
Constant values
3, 0.45
0-arity functions
rand, go left

3.2.1

Closure

For GP to work effectively, most function sets are required to have an important property known as closure (Koza, 1992), which can in turn be broken
down into the properties of type consistency and evaluation safety.
Type consistency is required because subtree crossover (as described in
Section 2.4) can mix and join nodes arbitrarily. As a result it is necessary
that any subtree can be used in any of the argument positions for every function in the function set, because it is always possible that subtree crossover
will generate that combination. It is thus common to require that all the
functions be type consistent, i.e., they all return values of the same type,
and that each of their arguments also have this type. For example +, -, *,
and / can can be defined so that they each take two integer arguments and
return an integer. Sometimes type consistency can be weakened somewhat
by providing an automatic conversion mechanism between types. We can,
for example, convert numbers to Booleans by treating all negative values as
false, and non-negative values as true. However, conversion mechanisms can
introduce unexpected biases into the search process, so they should be used
with care.
The type consistency requirement can seem quite limiting but often simple restructuring of the functions can resolve apparent problems. For example, an if function is often defined as taking three arguments: the test, the
value to return if the test evaluates to true and the value to return if the
test evaluates to false. The first of these three arguments is clearly Boolean,
which would suggest that if cant be used with numeric functions like +.

22

3 Getting Ready to Run Genetic Programming

This, however, can easily be worked around by providing a mechanism to

convert a numeric value into a Boolean automatically as discussed above.
Alternatively, one can replace the 3-input if with a function of four (numeric) arguments a, b, c, d. The 4-input if implements If a < b then return
value c otherwise return value d.
An alternative to requiring type consistency is to extend the GP system. Crossover and mutation might explicitly make use of type information
so that the children they produce do not contain illegal type mismatches.
When mutating a legal program, for example, mutation might be required
to generate a subtree which returns the same type as the subtree it has just
deleted. This is discussed further in Section 6.2.
The other component of closure is evaluation safety. Evaluation safety
is required because many commonly used functions can fail at run time. An
evolved expression might, for example, divide by 0, or call MOVE FORWARD
when facing a wall or precipice. This is typically dealt with by modifying
the normal behaviour of primitives. It is common to use protected versions
of numeric functions that can otherwise throw exceptions, such as division,
logarithm, exponential and square root. The protected version of a function
first tests for potential problems with its input(s) before executing the corresponding instruction; if a problem is spotted then some default value is
returned. Protected division (often notated with %) checks to see if its second
argument is 0. If so, % typically returns the value 1 (regardless of the value
of the first argument).1 Similarly, in a robotic application a MOVE AHEAD
instruction can be modified to do nothing if a forward move is illegal or if
moving the robot might damage it.
An alternative to protected functions is to trap run-time exceptions and
strongly reduce the fitness of programs that generate such errors. However,
if the likelihood of generating invalid expressions is very high, this can lead
to too many individuals in the population having nearly the same (very
poor) fitness. This makes it hard for selection to choose which individuals
might make good parents.
One type of run-time error that is more difficult to check for is numeric
overflow. If the underlying implementation system throws some sort of exception, then this can be handled either by protection or by penalising as
discussed above. However, it is common for implementation languages to
ignore integer overflow quietly and simply wrap around. If this is unacceptable, then the GP implementation must include appropriate checks to catch
and handle such overflows.

1 The decision to return the value 1 provides the GP system with a simple way to
generate the constant 1, via an expression of the form (% x x). This combined with a
similar mechanism for generating 0 via (- x x) ensures that GP can easily construct
these two important constants.

3.2 Step 2: Function Set

3.2.2

23

Sufficiency

There is one more property that primitives sets should have: sufficiency.
Sufficiency means it is possible to express a solution to the problem at hand
using the elements of the primitive set.2 Unfortunately, sufficiency can be
guaranteed only for those problems where theory, or experience with other
methods, tells us that a solution can be obtained by combining the elements
of the primitive set.
As an example of a sufficient primitive set consider {AND, OR, NOT, x1, x2,
..., xN}. It is always sufficient for Boolean induction problems, since it can
produce all Boolean functions of the variables x1, x2, ..., xN. An example
of insufficient set is {+, -, *, /, x, 0, 1, 2}, which is unable to represent
transcendental functions. The function exp(x), for example, is transcendental and therefore cannot be expressed as a rational function (basically, a
ratio of polynomials), and so cannot be represented exactly by any combination of {+, -, *, /, x, 0, 1, 2}. When a primitive set is insufficient, GP
can only develop programs that approximate the desired one. However, in
many cases such an approximation can be very close and good enough for
the users purpose. Adding a few unnecessary primitives in an attempt to
ensure sufficiency does not tend to slow down GP overmuch, although there
are cases where it can bias the system in unexpected ways.

3.2.3

Evolving Structures other than Programs

There are many problems where solutions cannot be directly cast as computer programs. For example, in many design problems the solution is an
artifact of some type: a bridge, a circuit, an antenna, a lens, etc. GP has
been applied to problems of this kind by using a trick: the primitive set is set
up so that the evolved programs construct solutions to the problem. This is
analogous to the process by which an egg grows into a chicken. For example,
if the goal is the automatic creation of an electronic controller for a plant,
the function set might include common components such as integrator,
differentiator, lead, lag, and gain, and the terminal set might contain
reference, signal, and plant output. Each of these primitives, when
executed, inserts the corresponding device into the controller being built.
If, on the other hand, the goal is to synthesise analogue electrical circuits,
the function set might include components such as transistors, capacitors,
resistors, etc. See Section 6.3 for more information on developmental GP
systems.

2 More formally, the primitive set is sufficient if the set of all the possible recursive
compositions of primitives includes at least one solution.

24

3.3

3 Getting Ready to Run Genetic Programming

Step 3: Fitness Function

The first two preparatory steps define the primitive set for GP, and therefore
indirectly define the search space GP will explore. This includes all the
programs that can be constructed by composing the primitives in all possible
ways. However, at this stage, we still do not know which elements or regions
of this search space are good. I.e., which regions of the search space include
programs that solve, or approximately solve, the problem. This is the task
of the fitness measure, which is our primary (and often sole) mechanism
for giving a high-level statement of the problems requirements to the GP
system. For example, suppose the goal is to get GP to synthesise an amplifier
automatically. Then the fitness function is the mechanism which tells GP
to synthesise a circuit that amplifies an incoming signal. (As opposed to
evolving a circuit that suppresses the low frequencies of an incoming signal,
or computes its square root, etc. etc.)
Fitness can be measured in many ways. For example, in terms of: the
amount of error between its output and the desired output; the amount
of time (fuel, money, etc.) required to bring a system to a desired target
state; the accuracy of the program in recognising patterns or classifying
objects; the payoff that a game-playing program produces; the compliance
of a structure with user-specified design criteria.
There is something unusual about the fitness functions used in GP that
differentiates them from those used in most other evolutionary algorithms.
Because the structures being evolved in GP are computer programs, fitness
evaluation normally requires executing all the programs in the population,
typically multiple times. While one can compile the GP programs that make
up the population, the overhead of building a compiler is usually substantial,
so it is much more common to use an interpreter to evaluate the evolved
programs.
Interpreting a program tree means executing the nodes in the tree in
an order that guarantees that nodes are not executed before the value of
their arguments (if any) is known. This is usually done by traversing the
tree recursively starting from the root node, and postponing the evaluation
of each node until the values of its children (arguments) are known. Other
orders, such as going from the leaves to the root, are possible. If none
of the primitives have side effects, the two orders are equivalent.3 This
depth-first recursive process is illustrated in Figure 3.1. Algorithm 3.1 gives
a pseudocode implementation of the interpretation procedure. The code
assumes that programs are represented as prefix-notation expressions and
that such expressions can be treated as lists of components.
3 Functional operations like addition dont depend on the order in which their arguments are evaluated. The order of side-effecting operations such as moving or turning a
robot, however, is obviously crucial.

3.3 Step 3: Fitness Function

25

x=-1
+

0 x

-2

-1

-3

0 x

Figure 3.1: Example interpretation of a syntax tree (the terminal x is a

variable and has a value of -1). The number to the right of each internal
node represents the result of evaluating the subtree root at that node.
procedure: eval( expr )
1: if expr is a list then
2:
proc = expr(1) {Non-terminal: extract root}
3:
if proc is a function then
4:
value = proc( eval(expr(2)), eval(expr(3)), ... ) {Function: evaluate
arguments}
5:
else
6:
value = proc( expr(2), expr(3), ...) {Macro: dont evaluate arguments}
7:
end if
8: else
9:
if expr is a variable or expr is a constant then
10:
value = expr {Terminal variable or constant: just read the value}
11:
else
12:
value = expr() {Terminal 0-arity function: execute}
13:
end if
14: end if
15: return value
Notes: expr is an expression in prefix notation, expr(1) represents the primitive at the root of the expression, expr(2) represents the first argument of
that primitive, expr(3) represents the second argument, etc.
Algorithm 3.1: Interpreter for genetic programming

26

3 Getting Ready to Run Genetic Programming

In some problems we are interested in the output produced by a program,

namely the value returned when we evaluate the tree starting at the root
node. In other problems we are interested in the actions performed by a
program composed of functions with side effects. In either case the fitness
of a program typically depends on the results produced by its execution on
many different inputs or under a variety of different conditions. For example
the program might be tested on all possible combinations of inputs x1, x2,
..., xN. Alternatively, a robot control program might be tested with the
robot in a number of starting locations. These different test cases typically
contribute to the fitness value of a program incrementally, and for this reason
are called fitness cases.
Another common feature of GP fitness measures is that, for many practical problems, they are multi-objective, i.e., they combine two or more different elements that are often in competition with one another. The area of
multi-objective optimisation is a complex and active area of research in GP
and machine learning in general. See Chapter 9 and also (Deb, 2001).

3.4

Step 4: GP Parameters

The fourth preparatory step specifies the control parameters for the run.
The most important control parameter is the population size. Other control
parameters include the probabilities of performing the genetic operations, the
maximum size for programs and other details of the run.
It is impossible to make general recommendations for setting optimal
parameter values, as these depend too much on the details of the application.
However, genetic programming is in practice robust, and it is likely that
many different parameter values will work. As a consequence, one need not
typically spend a long time tuning GP for it to work adequately.
It is common to create the initial population randomly using ramped
half-and-half (Section 2.2) with a depth range of 26. The initial tree sizes
will depend upon the number of the functions, the number of terminals
and the arities of the functions. However, evolution will quickly move the
population away from its initial distribution.
Traditionally, 90% of children are created by subtree crossover. However, the use of a 50-50 mixture of crossover and a variety of mutations (cf.
Chapter 5) also appears to work well.
In many cases, the main limitation on the population size is the time
taken to evaluate the fitnesses, not the space required to store the individuals. As a rule one prefers to have the largest population size that your
system can handle gracefully; normally, the population size should be at
least 500, and people often use much larger populations.4 Often, to a first
4 There

are, however, GP systems that frequently use much smaller populations. These

3.5 Step 5: Termination and solution designation

27

approximation, GP runtime can be estimated by the product of: the number

of runs R, the number of generations G, the size of the population P , the
average size of the programs s and the number of fitness cases F .
Typically, the number of generations is limited to between ten and fifty;
the most productive search is usually performed in those early generations,
and if a solution hasnt been found then, its unlikely to be found in a
reasonable amount of time. The folk wisdom on population size is to make
it as large as possible, but there are those who suggest using many runs with
much smaller populations instead. Some implementations do not require
arbitrary limits of tree size. Even so, because of bloat (the uncontrolled
growth of program sizes during GP runs; see Section 11.3), it is common to
impose either a size or a depth limit or both (see Section 11.3.2).
Sometimes the number of fitness cases is limited by the amount of training data5 available. In this case, the fitness function should use all of it.
(One does not necessarily need to use verification or holdout data, since
over-fitting can be avoided by other means, as discussed in Section 13.12,
page 140.) In other cases, e.g. 22-bit even parity, there can almost be too
much training data. Then the fitness function may be reduced to use just a
subset of the training data. This does not necessarily have to be done manually as there are a number of algorithms that dynamically change the test
set as the GP runs. (These and other speedup techniques will be discussed
in Chapter 13, particularly Section 10.1, page 83.)
It is common to record these details in a tableau, such as Table 4.1 on
page 31.

3.5

Step 5: Termination and solution designation

The fifth preparatory step consists of specifying the termination criterion

and the method of designating the result of the run. The termination criterion may include a maximum number of generations to be run as well as
a problem-specific success predicate. Typically, the single best-so-far individual is then harvested and designated as the result of the run, although
one might wish to return additional individuals and data as necessary or
appropriate for the problem domain.

typically rely more on mutation than crossover for their primary search mechanism.
5 Training data refers to the test cases used to evaluate the fitness of the evolved
individuals.

Chapter 4

Example
Genetic Programming Run
This chapter provides an illustrative run of GP in which the goal is to
automatically create a program with a target input/output behaviour. In
particular, we want to evolve an expression whose values match those of
the quadratic polynomial x2 + x + 1 in the range [1, +1]. The process of
mechanically creating a computer program that fit certain numerical data
is sometimes called system identification or symbolic regression (see Section 12.2 for more).
We begin with the five preparatory steps from the previous chapter and
then describe in detail the events in one run.

4.1

Preparatory Steps

The purpose of the first two preparatory steps is to specify the ingredients
the evolutionary process can use to construct potential solutions. Because
the problem is to find a mathematical function of one independent variable,
x, the terminal set (the inputs of the to-be-evolved programs) must include
this variable. The terminal set also includes ephemeral random constants
drawn from some reasonable range,1 say from 5.0 to +5.0, as described in

1 What is a reasonable range is likely to be extremely problem dependent. While in

theory you can build up large constants using small constants and arithmetic operators,
the performance of your system is likely to improve considerably if you provide constants
of roughly the right magnitude from the beginning. Your choice of genetic operators can
also be important here. If youre finding that your system is struggling to evolve the
right constants, it may be helpful to introduce mutation operators specifically designed
to search of the space of constants.

29

30

4 Example Genetic Programming Run

Section 3.1. Thus the terminal set, T , is

T = {x, <}.
The statement of the problem does not specify which functions may be
employed in the to-be-evolved program. One simple choice for the function
set is the four ordinary arithmetic functions: addition, subtraction, multiplication and division. Most numeric regression problems will require at
least these operations, sometimes with additional functions such as sin() and
log(). We will use the simple function set
F = {+, -, *, %},
where % is protected division as discussed in Section 3.2.1. Note that the
target polynomial can be expressed exactly using the terminal and function
sets we have chosen, so these primitives are sufficient (cf. page 23) for the
quadratic polynomial problem.
The third preparatory step involves constructing the fitness measure that
specifies what the user wants. The high-level goal of this problem is to find
a program whose output is equal to the values of the quadratic polynomial
x 2 +x +1. Therefore, the fitness assigned to a particular individual in the
population must reflect how closely the output of an individual program
comes to the target polynomial x2 + x + 1.
In principle, the fitness measure could be defined in terms of the mathematical integral of the difference between the evolved function and the target
function. However, for most symbolic regression problems, it is not practical
or even possible to compute the value of the integral analytically. Thus, it
is common to define the fitness to be the sum of absolute errors measured
at different values of the independent variable x in the range [1.0, +1.0].
In particular, we will measure the errors for x {1.0, 0.9, , 0.9, 1.0}.
A smaller value of fitness (error) is better; a fitness (error) of zero would
indicate a perfect fit. With this definition, our fitness is (approximately)
proportional to the area between the parabola x2 + x + 1 and the curve
representing the candidate individual (see Figure 4.2 for examples).
The fourth step is where we set our run parameters. The population size
in this small illustrative example will be just four. The population size for
a run of GP typically consists of thousands of individuals, but we will use
this tiny population size to keep the example manageable. The crossover
operation is commonly used to generate about 90% of the individuals in
the population; the reproduction operation (where a fit individual is simply copied from one generation to the next) is used to generate about 8%
of the population; the mutation operation is used to generate about 1% of
the population; and the architecture-altering operations (see Section 6.1.2)
are used to generate perhaps 1% of the population. However, because this
example involves an abnormally small population of only four individuals,

4.2 Step-by-Step Sample Run

31

Table 4.1: Parameters for example genetic programming run

Objective:
Find program whose output matches x2 + x + 1 over the
range 1 x +1.
Function set: +, , % (protected division), and ; all operating on floats
Terminal set: x, and constants chosen randomly between 5 and +5
Fitness:
sum of absolute errors for x {1.0, 0.9, . . . 0.9, 1.0}
Selection:
fitness proportionate (roulette wheel) non elitist
Initial pop:
ramped half-and-half (depth 1 to 2. 50% of terminals are
constants)
Parameters:
population size 4, 50% subtree crossover, 25% reproduction,
25% subtree mutation, no tree size limits
Termination: Individual with fitness better than 0.1 found

the crossover operation will be used twice (each time generating one individual), which corresponds to a crossover rate of 50%, while the mutation
and reproduction operations will each be used to generate one individual.
These are therefore applied with a rate of 25% each. For simplicity, the
architecture-altering operations are not used for this problem.
In the fifth and final step we need to specify a termination condition. A
reasonable termination criterion for this problem is that the run will continue
from generation to generation until the fitness (or error) of some individual
is less than 0.1. In this contrived example, our example run will (atypically)
yield an algebraically perfect solution with a fitness of zero after just one
generation.

4.2

Step-by-Step Sample Run

Now that we have performed the five preparatory steps, the run of GP can
be launched. The GP setup is summarised in Table 4.1.

4.2.1

Initialisation

GP starts by randomly creating a population of four individual computer

programs. The four programs are shown in Figure 4.1 in the form of trees.
The first randomly constructed program tree (Figure 4.1a) is equivalent
to the expression x+1. The second program (Figure 4.1b) adds the constant
terminal 1 to the result of multiplying x by x and is equivalent to x 2 +1. The
third program (Figure 4.1c) adds the constant terminal 2 to the constant
terminal 0 and is equivalent to the constant value 2. The fourth program
(Figure 4.1d) is equivalent to x.

32

4 Example Genetic Programming Run

(a)

(b)

1
x+1

+
1

(d)

(c)

x
2

x +1

-1
2

-2

Figure 4.1: Initial population of four randomly created individuals of generation 0.

4.2.2

Fitness Evaluation

Randomly created computer programs will typically be very poor at solving

any problem. However, even in a population of randomly created programs,
some programs are better than others. The four random individuals from
generation 0 in Figure 4.1 produce outputs that deviate by different amounts
from the target function x2 + x + 1. Figure 4.2 compares the plots of each of
the four individuals in Figure 4.1 and the target quadratic function x2 +x+1.
The sum of absolute errors for the straight line x +1 (the first individual) is
7.7 (Figure 4.2a). The sum of absolute errors for the parabola x 2 +1 (the
second individual) is 11.0 (Figure 4.2b). The sums of the absolute errors for
the remaining two individuals are 17.98 (Figure 4.2c) and 28.7 (Figure 4.2d).
As can be seen in Figure 4.2, the straight line x +1 (Figure 4.2a) is closer
to the parabola x2 +x+1 in the range from 1 to +1 than any of three other
programs in the population. This straight line is, of course, not equivalent to
the parabola x2 + x + 1; it is not even a quadratic function. It is merely the
best candidate that happened to emerge from the blind (and very limited)
random search of generation 0.
In the valley of the blind,
the one-eyed man is king.

4.2.3

Selection, Crossover and Mutation

After the fitness of each individual in the population is found, GP then

probabilistically selects the fitter programs from the population to act as
the parents of the next generation. The genetic operations are applied to
the selected individuals to create offspring programs. The important point
is that our selection process is not greedy. Individuals that are known to be
inferior still have some chance of being selected. The best individual in the
population is not guaranteed to be selected and the worst individual in the

4.2 Step-by-Step Sample Run

33

-1

-1

(a)

(b)

-2

-2

-1

-1

(c)
-2

(d)
-2

Figure 4.2: Graphs of the evolved functions from generation 0. The solid
line in each plot is the target function x2 + x + 1, with the dashed line
being the evolved functions from the first generation (see Figure 4.1). The
fitness of each of the four randomly created individuals of generation 0 is
approximately proportional to the area between two curves, with the actual
fitness values being 7.7, 11.0, 17.98 and 28.7 for individuals (a) through (d),
respectively.

population will not necessarily be excluded.

In this example we will start with the reproduction operation. Because
the first individual (Figure 4.1a) is the most fit individual in the population,
it is very likely to be selected to participate in a genetic operation. Let us
suppose that this particular individual is, in fact, selected for reproduction.
If so, it is copied, without alteration, into the next generation (generation 1).
It is shown in Figure 4.3a as part of the population of the new generation.
We next perform the mutation operation. Because selection is probabilistic, it is possible that the third best individual in the population (Figure 4.1c) is selected. One of the three nodes of this individual is then randomly picked as the site for the mutation. In this example, the constant
terminal 2 is picked as the mutation site. This program is then randomly
mutated by deleting the entire subtree rooted at the picked point (in this
case, just the constant terminal 2) and inserting a subtree that is randomly
constructed in the same way that the individuals of the initial random population were originally created. In this particular instance, the randomly
grown subtree computes x divided by x using the protected division oper-

34

4 Example Genetic Programming Run

(a)

(b)

(c)

(d)

x
x+1

x + x+1

Figure 4.3: Population of generation 1 (after one reproduction, one mutation, and two one-offspring crossover operations).

ation %. The resulting individual is shown in Figure 4.3b. This particular

mutation changes the original individual from one having a constant value
of 2 into one having a constant value of 1, improving its fitness from 17.98
to 11.0.
Finally, we use the crossover operation to generate our final two individuals for the next generation. Because the first and second individuals in
generation 0 are both relatively fit, they are likely to be selected to participate in crossover. However, selection can always pick suboptimal individuals. So, let us assume that in our first application of crossover the pair of
selected parents is composed of the above-average tree in Figures 4.1a and
the below-average tree in Figure 4.1d. One point of the first parent, namely
the + function in Figure 4.1a, is randomly picked as the crossover point for
the first parent. One point of the second parent, namely the leftmost terminal x in Figure 4.1d, is randomly picked as the crossover point for the second
parent. The crossover operation is then performed on the two parents. The
offspring (Figure 4.3c) is equivalent to x and is not particularly noteworthy.
Let us now assume, that in our second application of crossover, selection
chooses the two most fit individuals as parents: the individual in Figure 4.1b
as the first parent, and the individual in Figure 4.1a as the second. Let us
further imagine that crossover picks the leftmost terminal x in Figure 4.1b
as a crossover point for the first parent, and the + function in Figure 4.1a as
the crossover point for the second parent. Now the offspring (Figure 4.3d)
is equivalent to x2 + x + 1 and has a fitness (sum of absolute errors) of zero.

4.2 Step-by-Step Sample Run

4.2.4

35

Termination and Solution Designation

Because the fitness of the individual in Figure 4.3d is below 0.1, the termination criterion for the run is satisfied and the run is automatically terminated.
This best-so-far individual (Figure 4.3d) is then designated as the result of
the run.
Note that the best-of-run individual (Figure 4.3d) incorporates a good
trait (the quadratic term x 2 ) from the first parent (Figure 4.1b) with two
other good traits (the linear term x and constant term of 1) from the second
parent (Figure 4.1a). The crossover operation thus produced a solution to
this problem by recombining good traits from these two relatively fit parents
into a superior (indeed, perfect) offspring.
This is, obviously, a highly simplified example, and the dynamics of a
real GP run are typically far more complex than what is presented here.
Also, in general, there is no guarantee that an exact solution like this will
be found by GP.

Part II

Advanced Genetic
Programming

In which a search is organdized . . .

and Piglet encounters the Heffalump of Bloat.

37

Chapter 5

Alternative Initialisations
and Operators in
Tree-based GP
The genetic programming system described in the preceding chapters is just
the beginning; in many ways it is the simplest thing that could possibly
work. Most of the techniques described in Part I date back to the late
1980s and early 1990s, a wide array of alternatives and extensions have
been explored since. A full catalogue of these would be far beyond the
scope of this book. The chapters in Part II survey a number of the more
prominent or historically important extensions to GP, particularly (but not
exclusively) in relation to the tree-based representation for programs.
We start, in this chapter, by reviewing a variety of initialisation strategies
(Section 5.1) and genetic operators (Sections 5.2 and 5.3) for tree-based GP
not covered in Part I. We also briefly look at some hybridisations of GP
with other techniques (Section 5.4).

5.1

Constructing the Initial Population

Kozas ramped half-and-half method is the most common way of creating the
initial GP population (cf. Section 2.2, page 11). However, there are several
other ways of constructing a collection of random trees. In Section 5.1.2
we will briefly consider an unexpected impact of population initialisation.
There has also been some work with non-random or informed starting points
(cf. Section 5.1.3).
39

40

5.1.1

5 Alternative Initialisations and Operators in Tree-based GP

Uniform Initialisation

The shape of the initial trees can be lost within a few generations (more on
this below). However, a good start given by the initial population can still be
crucial to the success of a GP run. In general, there are an infinite number of
possible computer programs. This means that it is impossible to search them
uniformly. Therefore, any method used to create the initial population will
have a bias. For example, ramped half-and-half tends to create bushy trees.
Such trees have a higher proportion of solutions to symmetric problems,
such as parity. Conversely, the smallest solution to the Sante Fe ant trailfollowing problem is more randomly shaped (Langdon and Poli, 1998a). This
is partly why ramped half-and-half is very poor at finding programs which
can navigate the Sante Fe trail. Another reason is that many of the programs
generated by ramped half-and-half (with standard parameters) are simply
too small. Chellapilla (1997a) claims good results when the size of the initial
trees was more tightly controlled.
Iba (1996a) and Bohm and Geyer-Schulz (1996) report methods to precisely sample trees uniformly based on Alonsos bijective algorithm (Alonso
and Schott, 1995). Although this algorithm has been criticised (Luke,
2000) for being computationally expensive, it can be readily used in practice. Langdon (2000) introduced the ramped uniform initialisation which
extends Alonsos bijective algorithm by allowing the user to specify a range
of initial tree sizes. It then generates equal numbers of random trees
for each length in the chosen range. (C++ code can be obtained from
ftp://cs.ucl.ac.uk/genetic/gp-code/rand tree.cc.)
With these more uniform initialisations, most trees are asymmetric
with some leaves very close to the root of the tree. This is quite different
from the trees generated by ramped half-and-half which are on average some
distance from the root. Uniform sampling may be better in problems where
the desired solutions are asymmetric with some leaves being much more
important than others. For example, in data mining it is common to look
for solutions with a few dominant variables (which may be close to the root
node) whilst other variables are of little or no interest and may be some
distance from the root (or indeed not present in the tree). On the other
hand, problems like multiplexer or parity require all the inputs to be used
and are of similar importance. Bushier trees may be better at solving such
problems.

5.1.2

Initialisation may Affect Bloat

Crossover has a strong preference for creating a very non-uniform distributions of tree sizes (Poli, Langdon, and Dignum, 2007). Crossover generates very short programs much more often than longer ones. Selection can
only partially combat this tendency. Typically, crossover will totally rear-

5.1 Constructing the Initial Population

41

range the size and shape of the initial trees within a few generations. As
discussed in Section 11.3.1 (page 101), the excessive sampling of short programs appears to be an important cause of bloat (the uncontrolled growth
of programs during GP runs, which will be described in more detail in Section 11.3, page 101 onwards). It has been shown (Dignum and Poli, 2007)
that when the initial population is created with the size distribution preferred by crossover (see Section 11.3.1), bloat is more marked. The distribution has a known mathematical formula (it is a Lagrange distribution of
the second kind), but in practice it can be created by simply performing
multiple rounds of crossover on a population created in the traditional way
before the GP run starts. This is known as Lagrange initialisation. These
findings suggest that initialisation methods which tend to produce many
short programs may in fact induce bloat sooner than methods that produce
distributions more skewed towards larger programs.

5.1.3

Seeding

The most common way of starting a GP run from an informed non-random

point is seeding the initial population with an individual which, albeit not
a solution, is thought to be a good starting point. Such a seed may have
been produced by an earlier GP run or perhaps constructed by the user
(Aler, Borrajo, and Isasi, 2002; Holmes, 1995; Hsu and Gustafson, 2001;
Langdon and Nordin, 2000; Langdon and Treleaven, 1997; Westerberg and
Levine, 2001). However, Marek, Smart, and Martin (2002) reported that
hand written programs may not be robust enough to prosper in an evolving
population.
One point to be careful of is that such a seed individual is liable to be
much better than randomly created trees. Thus, its descendants may take
over the population within a few generations. So, under evolution the seeded
population is initially liable to lose diversity rapidly. Furthermore, depending upon the details of the selection scheme used, a single seed individual
may have some chance of being removed from the population. Both problems
are normally dealt with by filling the whole population with either identical
or mutated copies of the seed. This method creates a low diversity initial
population in a controlled way, thereby avoiding the initial uncontrolled loss
of diversity associated with single seeds. Furthermore, with many copies of
the seed, few selection methods will have much chance of removing all copies
of the seed before they are able to create children. Diversity preserving techniques, such as multi-objective GP (e.g., (Parrott, Li, and Ciesielski, 2005),
(Setzkorn, 2005) and Chapter 9), demes (Langdon, 1998) (see Section 10.3),
fitness sharing (Goldberg, 1989) and the use of multiple seed trees, might
also be good cures for the problems associated with the use of a single seed.
In any case, the diversity of the population should be monitored to ensure
that there is significant mixing of different initial trees.

42

5.2
5.2.1

5 Alternative Initialisations and Operators in Tree-based GP

GP Mutation
Is Mutation Necessary?

Mutation was used in early experiments in the evolution of programs, e.g.,

in (Bickel and Bickel, 1987; Cramer, 1985; Fujiki and Dickinson, 1987). It
was not, however, used in (Koza, 1992) and (Koza, 1994), as Koza wished to
demonstrate that mutation was not necessary and that GP was not performing a simple random search. This has significantly influenced the field, and
mutation is often omitted from GP runs. While mutation is not necessary
for GP to solve many problems, OReilly (1995) argued that mutation in
combination with simulated annealing or stochastic iterated hill climbing
can perform as well as crossover-based GP in some cases. Nowadays, mutation is widely used in GP, especially in modelling applications. Koza also
advises to use of a low level of mutation; see, for example, (Koza, Bennett,
Andre, and Keane, 1996b).
Comparisons of crossover and mutation suggest that including mutation
can be advantageous. Chellapilla (1997b) found that a combination of six
mutation operators performed better than previously published GP work on
four simple problems. Harries and Smith (1997) also found that mutation
based hill climbers outperformed crossover-based GP systems on similar
problems. Luke and Spector (1997) suggested that the situation is complex,
and that the relative performance of crossover and mutation depends on
both the problem and the details of the GP system.

5.2.2

Mutation Cookbook

With linear bit string GAs, mutation usually consists of random changes in
bit values. In contrast, in GP there are many mutation operators in use.
Often multiple types of mutation are beneficially used simultaneously (e.g.,
see (Kraft, Petry, Buckles, and Sadasivan, 1994) and (Angeline, 1996)). We
describe a selection of mutation operators below:
Subtree mutation replaces a randomly selected subtree with another randomly created subtree (Koza, 1992, page 106). Kinnear (1993) defined
a similar mutation operator, but with a restriction that prevents the
offspring from being more than 15% deeper than its parent.
Size-fair subtree mutation was proposed in two forms by Langdon
(1998). In both cases, the new random subtree is, on average, the
same size as the code it replaces. The size of the random code is given
either by the size of another random subtree in the program or chosen
at random in the range [l/2, 3l/2] (where l is the size of the subtree
being replaced). The first of these methods samples uniformly in the
space of possible programs, whereas the second samples uniformly in

5.2 GP Mutation

43

the space of program lengths. Experiments suggested that there was

far more bloat (cf. Section 11.3.1 page 101) with the first mutation
operator.
Node replacement mutation (also known as point mutation) is similar
to bit string mutation in that it randomly changes a point in the
individual. In linear GAs the change would be a bit flip. In GP,
instead, a node in the tree is randomly selected and randomly changed.
To ensure the tree remains legal, the replacement node has the same
number of arguments as the node it is replacing, e.g. (McKay, Willis,
and Barton, 1995, page 488).
Hoist mutation creates a new offspring individual which is copy of a randomly chosen subtree of the parent. Thus, the offspring will be smaller
than the parent and will have a different root node (Kinnear, 1994a).
Shrink mutation replaces a randomly chosen subtree with a randomly
created terminal (Angeline, 1996). This is a special case of subtree
mutation where the replacement tree is a terminal. As with hoist
mutation, it is motivated by the desire to reduce program size.
Permutation mutation selects a random function node in a tree and then
randomly permuting its arguments (subtrees). Koza (1992) used permutation in one experiment [page 600] where it was shown to have
little effect. In contrast, Maxwell (1996) had more success with a mutation operator called swap, which is simply a permutation mutation
restricted to binary non-commutative functions.
Mutating constants at random Schoenauer, Sebag, Jouve, Lamy, and
Maitournam (1996) mutated constants by adding random noise from
a Gaussian distribution. Each change to a constant was considered a
separate mutation.
Mutating constants systematically A variety of potentially expensive
optimisation tools have been applied to try and fine-tune an existing
program by finding the best value for the constants within it. Indeed
STROGANOFF (Iba, Sato, and de Garis, 1995b; Nikolaev and Iba,
2006) optimises each tree modified by crossover. Clever mechanisms
are employed to minimise the computation required.
(McKay et al., 1995, page 489) is more in keeping with traditional GP
and uses a mutation operator that operates on terminals, replacing input variables by constants and vice versa. In this approach whenever
a new constant is introduced [. . . ] a non-linear least squares optimisation is performed to obtain the best value of the constant(s). Schoenauer, Lamy, and Jouve (1995) also used a mutation operator that

44

5 Alternative Initialisations and Operators in Tree-based GP

affects all constants in an individual where a numerical partial gradient ascent is achieved to reach the nearest local optimum. Finally,
Sharman, Esparcia Alcazar, and Li (1995) used simulated annealing to
update numerical values (which represented signal amplification gains)
within individuals.

5.3

GP Crossover

During biological sexual reproduction, the genetic material from both

mother and father is combined in such a way that genes in the child are
in approximately the same position as they were in its parents. This is quite
different from traditional tree-based GP crossover, which can move a subtree
to a totally different position in the tree structure.
Crossover operators that tend to preserve the position of genetic material are called homologous, and several notions of homologous crossover
have been proposed for GP. It is fairly straightforward to realise homologous crossover when using linear representations, and homologous operators
are widely used in linear GP (cf. Figure 7.4, page 65) (Defoin Platel, Clergue,
and Collard, 2003; Francone, Conrads, Banzhaf, and Nordin, 1999; Hansen,
2003; Hansen, Lowry, Meservy, and McDonald, 2007; Nordin, Banzhaf, and
Francone, 1999; ONeill, Ryan, Keijzer, and Cattolico, 2003). Various forms
of homologous crossover have also been proposed for tree-based GP (Collet, 2007; Langdon, 2000; Lones, 2003; MacCallum, 2003; Yamamoto and
Tschudin, 2005).
The oldest homologous crossover in tree-based GP is one-point crossover
(Langdon and Poli, 2002; Poli and Langdon, 1997, 1998a). This works by selecting a common crossover point in the parent programs and then swapping
the corresponding subtrees. To allow for the two parents having different
shapes, one-point crossover analyses the two trees from the root nodes and
selects the crossover point only from the parts of the two trees in the common
region (see Figure 5.1). In the common region, the parents have the same
shape.1 The common region is related to homology, in the sense that the
common region represents the result of a matching process between parent
trees. Within the common region between two parent trees, the transfer of
homologous primitives can happen like it does in a linear bit string genetic
algorithm.
Uniform crossover for trees (Poli and Langdon, 1998b) works (in the
common region) like uniform crossover in GAs. That is, the offspring are
created by visiting the nodes in the common region and flipping a coin at
1 Nodes in the common region need not be identical but they must have the same
arity. That is, they must both be leaves or both be functions with the same number of
inputs.

5.3 GP Crossover

45

Parent 1

Parent 2

*
+

sin
x

*
x y

*
y

Alignment

sin
x

*
y

*
y

Parent 2
x*

Swap

y xy

Selection of
Common
Crossover Point

Common
Region

+*

Offspring 2

sin*
y x

*
y

Offspring 1

+*

Parent 1

Crossover Point

+*

Parent 1 sin*
y x

+ * Parent 2
y xy

x*
y

Figure 5.1: Example of one-point crossover between parents of different

sizes and shapes.

each locus to decide whether the corresponding offspring node should be

picked from the first or the second parent. If a node to be inherited belongs
to the base of the common region and is a function, then the subtree rooted
there is inherited as well. With this form of crossover, there can be a greater
mixing of the code near the root than with other operators.
In context-preserving crossover (Dhaeseleer, 1994), the crossover points
are constrained to have the same coordinates, like in one-point crossover.
Note that the crossover points are not limited to the common region.
In size-fair crossover (Langdon, 1999a, 2000) the first crossover point is
selected randomly, as with standard crossover. Then the size of the subtree
to be removed from the first parent is calculated. This is used to constrain
the choice of the second crossover point so as to guarantee that the subtree
excised from the second parent will not be unfairly big.
Harries and Smith (1997) suggested five new crossover operators that
are like standard crossover but with probabilistic restrictions on the depth
of crossover points within the parent trees.
Since crossover and mutation are specific to the representation used in
GP, each new representation tends to need new crossover and mutation
operators. For example ripple crossover (ONeill et al., 2003) is a way
of looking at crossover in grammatical evolution (Section 6.2.3 page 55).

46

5 Alternative Initialisations and Operators in Tree-based GP

As we shall see in Chapter 7, specific crossover operators exist for linear GP (Section 7.1) and graph based GP systems (Section 7.2), such as
PDGP (page 65), PADO (page 67) and Cartesian GP (page 67).

5.4

Other Techniques

GP can be hybridised with other techniques. For example, Iba, de Garis, and
Sato (1994), Nikolaev and Iba (2006), and Zhang and M
uhlenbein (1995)
have incorporated information theoretic and minimum description length
ideas into GP fitness functions to provide a degree of regularisation and
so avoid over-fitting (and bloat, see Section 11.3). As mentioned in Section 6.2.3, computer language grammars can be incorporated into GP.
Whereas genetic programming typically uses an evolutionary algorithm
to search the space of computer programs, various other heuristic search
methods can also be applied to program search, including: enumeration
(Olsson, 1995), hill climbing (OReilly and Oppacher, 1994a), and simulated annealing (OReilly, 1996; Tsoulos and Lagaris, 2006). As discussed
in Chapter 8, it is also possible to extend Estimation of Distribution Algorithms (EDAs) to the variable size representations used in GP.
Another alternative is to use co-evolution with multiple populations,
where the fitness of individuals in one population depends on the behaviour
of individuals in other populations. There have been many successful applications of co-evolution in GP, including (Azaria and Sipper, 2005a; Brameier,
Haan, Krings, and MacCallum, 2006; Buason, Bergfeldt, and Ziemke, 2005;
Channon, 2006; Dolinsky, Jenkinson, and Colquhoun, 2007; Funes, Sklar,
Juille, and Pollack, 1998a; Gagne and Parizeau, 2007; Hillis, 1992; Hornby
and Pollack, 2001; Mendes, de B. Voznika, Nievola, and Freitas, 2001; Piaseczny, Suzuki, and Sawai, 2004; Schmidt and Lipson, 2006; Sharabi and
Sipper, 2006; Soule, 2003; Soule and Komireddy, 2006; Spector, 2002; Spector and Klein, 2006; Spector, Klein, Perry, and Feinstein, 2005b; Wilson and
Heywood, 2007; Zhang and Cho, 1999).
Finally, it is worth mentioning that program trees can be manipulated
with editing operations (Koza, 1992). For example, if the root node of
a subtree is but one of its arguments is always guaranteed to evaluate
to 0, then we can replace the subtree rooted there with the terminal 0.
If the root node of a subtree is + and one argument evaluates to 0, we
can replace the subtree with the other argument of the +. Editing can
reduce the complexity of evolved solutions and can make them easier to
understand. However, it may also lead to GP getting stuck in local optima,
so editing operations should probably be used sparingly at run time. Other
reorganisation operations of various types are also possible. For example,
after trees are generated by GP, (Garcia-Almanza and Tsang, 2006, 2007)
prune branches and combine branches from different trees.

Chapter 6

Modular, Grammatical
and Developmental
Tree-based GP
This chapter discusses advanced techniques that are primarily focused on
two important issues in genetic programming: modularity and constraint.
In Section 6.1 we explore the evolution of modular, hierarchical structures,
and in Section 6.2 we looks at ways of constraining the evolutionary process,
typically based on some sort of domain knowledge. We also look at using
GP to evolve programs which themselves develop solutions (Section 6.3) or
even construct other programs (Section 6.4).

6.1

Evolving Modular and

Hierarchical Structures

The construction of any highly complex object or individual, whether an oak

tree or an airliner, typically uses hierarchical, modular structures to manage
and organise that complexity. Animals develop in a highly regular way
that yields a hierarchical structure of components ranging from systems and
organs down to cells and organelles. GP, as described so far, is typically used
to evolve expressions that, while being suitable solutions to many problems,
rarely exhibit any large-scale modular structure.
Given the pervasiveness of hierarchical structure as an organisational tool
in both biology and engineering, it seems likely that such modular structure
could be valuable in genetic programming as well. Consequently, this has
been a subject of study from the early days of genetic programming. For
example, Angeline and Pollack (1992) created dynamic libraries of subtrees
47

48

6 Modular, Grammatical and Developmental Tree-based GP

taken from parts of fit GP trees. Special mutation operations allowed the
GP population to share code by referring to the same code within the library. Subsequently, Angeline suggested that the schemes advantages lay
in allowing GP individuals to access far more code than they actually held
within themselves, rather than principally in developing more modular code.
Rosca and Ballard (1996a) used a similar scheme, but were able to use much
more information from the fitness function to guide the selection of the code
to be inserted into the library and its subsequent use by members of the GP
population. Olsson (1999, 1995) later developed an abstraction operator for
use in his ADATE system, where sub-functions (anonymous lambda expressions) were automatically extracted. Unlike Angelines library approach,
Olssons modules remained attached to the individual they were extracted
from.
Kozas automatically defined functions (ADFs) (Koza, 1994) remain the
most widely used method of evolving reusable components and have been
used successfully in a variety of settings. Basic ADFs (covered in Section 6.1.1) use a fixed architecture specified in advance by the user. Koza
later extended this using architecture altering operations (Section 6.1.2),
which allow the architecture to evolve along with the programs.

6.1.1

Automatically Defined Functions

Human programmers organise sequences of repeated steps into reusable components such as subroutines, functions and classes. They then repeatedly
invoke these components, typically with different inputs. Reuse eliminates
the need to reinvent the wheel every time a particular sequence of steps
is needed. Reuse also makes it possible to exploit a problems modularities,
symmetries and regularities (thereby potentially accelerate the problemsolving process). This can be taken further, as programmers typically organise these components into hierarchies in which top level components call
lower level ones, which call still lower levels, etc. Kozas ADFs provide a
mechanism by which the evolutionary process can evolve these kinds of potentially reusable components. We will review the basic concepts here, but
ADFs are discussed in great detail in (Koza, 1994).
When ADFs are used, a program consists of multiple components. These
typically consist of one or more function-defining branches (i.e., ADFs), as
well as one or more main result-producing branches (the RPB ), as illustrated
in the example in Figure 6.1. The RPB is the main program that is
executed when the individual is evaluated. It can, however, call the ADFs,
which can in turn potentially call each other. A single ADF may be called
multiple times by the same RPB, or by a combination of the RPB and other
ADFs, allowing the logic that evolution has assembled in that ADF to be
re-used in different contexts.

6.1 Evolving Modular and Hierarchical Structures

49

ROOT

ADF1

ADF2

RPB

Figure 6.1: Example of program structure with two automatically-defined

functions (ADF1 and ADF2) and one result-producing branch (RPB).
Consider, for example, the following individual consisting of a resultproducing branch and a single ADF:
RPB : ADF(ADF(ADF(x)))
ADF : arg0 arg0

(6.1)
(6.2)

The ADF (Equation 6.2) is simply the squaring function, but by combining
this multiple times in the RPB (Equation 6.1) this individual computes x8
in a highly compact fashion.
It is important to not be fooled by a tidy example like this. ADFs
evolved in real applications are typically complex and can be very difficult
to understand. Further, simply including ADFs provides no guarantee of
modular re-use. As is discussed in Chapter 13, there are no silver bullets.
It may be that the RPB never calls an ADF or only calls it once. It is also
common for an ADF to not actually encapsulate any significant logic. For
example, an ADF might be as simple as a single terminal, in which case it
is essentially just providing a new name for that terminal.
In Kozas approach, each ADF is attached (as a branch) to a specific individual in the population. This is in contrast to both Angelines and Roscas
systems mentioned above, both of which have general pools of modules or
components which are shared across the population. Sometimes recursion
is allowed in ADFs, but this frequently leads to infinite computations. Typically, recursion is prevented by imposing an order on the ADFs within an
individual and by restricting calls so that ADFi can only call ADFj if i < j.
In the presence of ADFs, recombination operators are typically constrained to respect the larger structure. That is, during crossover, a subtree

50

6 Modular, Grammatical and Developmental Tree-based GP

from ADFi can only be swapped with a subtree from another individuals
ADFi .
The programs result-producing branch and its ADFs typically have different function and terminal sets. For example, the terminal set for ADFs
usually include arguments, such as arg0, arg1. Typically the user must
decide in advance the primitive sets, the number of ADFs and any call restrictions to prevent recursion. However, these choices can be evolved using
the architecture-altering operations described in Section 6.1.2.
Koza also proposed other types of automatically evolved program components (Koza, Andre, Bennet, and Keane, 1999). Automatically defined
iterations (ADIs), automatically defined loops (ADLs) and automatically
defined recursions (ADRs) provide means to reuse code. Automatically defined stores (ADSs) provide means to reuse the result of executing code.

6.1.2

Program Architecture and Architecture-Altering

Operations

Koza (1994) defined the architecture of a program to be the total number of

trees, the type of each tree (e.g., RPB, ADF, ADI, ADL, ADR, or ADS), the
number of arguments (if any) possessed by each tree, and, finally, if there is
more than one tree, the nature of the hierarchical references (if any) allowed
among the trees (e.g., whether ADF1 can call ADF2).
There are three ways to determine the architecture of the computer programs that will be evolved:
1. The user may specify in advance the architecture of the overall program, i.e., perform an architecture-defining preparatory step in addition to the five steps itemised in Chapter 3.
2. A run of genetic programming may employ the evolutionary design
of the architecture (as described in (Koza, 1994)), thereby enabling
the architecture of the overall program to emerge from a competitive
process during the run.
3. The run may employ a set of architecture-altering operations (Koza,
1994, 1995; Koza, Bennett, Andre, and Keane, 1999) which can create new ADFs, remove ADFs, and increase or decrease the number
of inputs an ADF has. Note that many architecture changes (such
as those defined in (Koza, 1994)) are designed not to initially change
the semantics of the program and, so, the altered program often has
exactly the same fitness as its parent. Nevertheless, the new arrangement of ADFs may make it easier for subsequent changes to evolve
better programs later.

6.2 Constraining Structures

51

Koza and his colleagues have used these architecture altering operations
quite widely in their genetic design work, where they evolve GP trees that
encode a collection of developmental operations that, when interpreted, generate a complex structure like a circuit or an optical system (see, for example,
Section 12.3, page 118).
The idea of architecture altering operations was extended to the extremely general Genetic Programming Problem Solver (GPPS), which is
described in detail in (Koza et al., 1999, part 4). This is an open ended
system which combines a small set of basic vector-based primitives with the
architecture altering operations in a way that can, in theory, solve a wide
range of problems with almost no input required from the user other than
the fitness function. The problem is that this open-ended system needs a
very carefully constructed fitness function to guide it to a viable solution, an
enormous amount of computational effort, or both. As a result it is currently
an idea of more conceptual than practical value.

6.2

Constraining Structures

As discussed in Section 3.2.1, most GP systems require type consistency

where all subtrees return data of the same type. This ensures that the output of any subtree can be used as one of the inputs to any node. The basic
subtree crossover operator shuffles tree components entirely randomly. Universal type compatibility ensures that crossover cannot lead to incompatible
connections between nodes. This is also required to stop mutation from
producing illegal programs.
An implicit assumption underlying this approach is that all combinations
of structures are equally likely to be useful. In many cases, however, we know
in advance that there are constraints on the structure of the solution, or we
have strong suspicions about the likely form solutions will take. In this
section, we will look at several different systems that use tools such as types
and grammars to bias or constrain search with the primary aim of increasing
the chance of finding a suitable program.
A problem domain might be naturally represented with multiple types.
This suggests that the functions used by GP and their arguments will not
necessarily be all of the same type. This can often be addressed through
creative definitions of functions and implicit type conversion. For example,
the Odin system (Holmes and Barclay, 1996) defines operations on inappropriate types to return a new fail object. These are handled by introducing
a binary fatbar that returns its first argument unless it is fail, in which case
it returns its second argument.
This sort of approach may not always be desirable. For example, if a
key goal is to evolve solutions that can be easily understood or analysed,
then one might prefer a GP system that is constrained structurally or via

52

6 Modular, Grammatical and Developmental Tree-based GP

a type system, since these often generate results that are more comprehensible (Haynes, Wainwright, Sen, and Schoenefeld, 1995), (Langdon, 1998,
page 126). Similarly, if there is domain knowledge that strongly suggests a
particular syntactic constraint on the solution, then ignoring that constraint
may make it much harder to find a solution.
We will focus here on three different approaches to constraining the syntax of the evolved expression trees in GP: simple structure enforcement
(Section 6.2.1), strongly typed GP (Section 6.2.2) and grammar-based constraints (Section 6.2.3). Finally, we consider the advantages and disadvantages of syntactic and type constraints and their biases (Section 6.2.4).

6.2.1

Enforcing Particular Structures

If a particular structure is believed or known to be important then one

can modify the GP system to require that all individuals have that structure (Koza, 1992). For example, if a problem is believed to have (or require)
a periodic solution, one might want to consider constraining the search to
solutions of the form a sin(b t). By allowing a and b to evolve freely
but keeping the rest of the structure fixed, one could restrict GP to evolving
expressions that are periodic. Syntax restrictions can also be used to make
GP follow sensible engineering practices. For example, we might want to
ensure that loop control variables appear in the correct parts of for loops
and nowhere else (Langdon, 1998, page 126).
Enforcing a user specified structure on the evolved solutions can be implemented in a number of ways. One could ensure that all the initial individuals
have the structure of interest (for example, generating random subtrees for
a and b while fixing the rest) and then constrain crossover and mutation
so that they do not alter any of the fixed regions of a tree. An alternative
approach would be to evolve the various (sub)components separately. One
could evolve pairs of trees (a, b) (like ADFs). Alternatively, one could have
two separate populations, one of which is used to evolve candidates for a
while the other is evolving candidates for b.
A form of constraint-directed search in GP was also proposed in (Tsang
and Jin, 2006; Tsang and Li, 2002) to help GP to focus on more promising
areas of the space.

6.2.2

Strongly Typed GP

Since constraints are often driven by or expressed using a type system, a

natural approach is to incorporate types and their constraints into the GP
system (Montana, 1995). In strongly typed GP, every terminal has a type,
and every function has types for each of its arguments and a type for its
return value. The process that generates the initial, random expressions,

6.2 Constraining Structures

53

and all the genetic operators are implemented so as to ensure that they do
not violate the type systems constraints.
Returning to the if example from Section 3.2.1 (page 21), we might have
an application with both numeric and Boolean terminals (e.g., get speed
and is food ahead). We might then have an if function that takes three
arguments: a test (Boolean), the value to return if the test is true, and
the value to return if the test is false. Assuming that the second and third
values are numbers, then the output of the if is also going to be numeric.
If we choose the test argument as a crossover point in the first parent, then
the subtree (excised from the second parent) to insert must have a Boolean
output. That is, we must find either a function which returns a Boolean or
a Boolean terminal in the other parent tree to be the root of the subtree
which we will insert into the new child. Conversely if we choose either the
second or third argument as a crossover point in the first parent, then the
inserted subtree must be numeric. In all three cases, given that both parents
are type correct, restricting the second crossover point in this way ensures
the child will also be type correct.
This basic approach to types can be extended to more complex type systems including simple generics (Montana, 1995), multi-level type systems
(Haynes, Schoenefeld, and Wainwright, 1996), fully polymorphic types (Olsson, 1994), and polymorphic higher-order type systems (Yu, 2001).

6.2.3

Grammar-based Constraints

Another natural way to express constraints is via grammars, and these have
been used in GP in a variety of ways (Gruau, 1996; Hoai, McKay, and
Abbass, 2003; ONeill and Ryan, 2003; Whigham, 1996; Wong and Leung,
1996). Many of these simply use a grammar as a means of expressing the
kinds of constraints discussed above in Section 6.2.1. For example, one could
enforce the structure for the period function using a grammar such as the
following:
tree
E
op
var

::=
::=
::=
::=

E sin(E t)
var | (E op E)
+ | | |
x | y | z

(6.3)

Each line in this grammar is known as a rewrite rule or a production rule.

Elements that cannot be rewritten are known as the terminals of the grammar 1 while symbols that appear on the left-hand-side of a rule are known
as non-terminal symbols.
1 Not

to be confused with the terminals in the primitive set of a GP system.

54

6 Modular, Grammatical and Developmental Tree-based GP

tree

var

sin

op

var

var

Figure 6.2: Example individual (a derivation tree) that might be evolved

in Whighams grammar-based GP system (Whigham, 1996) if the grammar
in Equation (6.3) was used. Rectangles represent non-terminal symbols of
the grammar.
In this sort of system, the grammar is typically used to ensure the initial
population is made up of legal grammatical programs. The grammar is
also used to guide the operations of the genetic operators. Thus we need to
keep track not only of the program itself, but also the syntax rules used to
derive it.
What actually is evolved in a grammar-based GP system depends on the
particular system. Whigham (1996), for example, evolved derivation trees,
which effectively are a hierarchical representation of which rewrite rules must
be applied, and in which order, to obtain a particular program. Figure 6.2
shows an example of a derivation tree representing a grammatical program
with respect to the grammar in Equation (6.3). In this system, crossover is
restricted to only swapping subtrees deriving from a common non-terminal
symbol in the grammar. So, for example, a subtree rooted by an E node
could be replaced by another also rooted by an E, while an E-rooted subtree
could not be replaced by an op-rooted one.
The actual program represented by a derivation tree can be obtained by
reading out the leaves of the tree one by one from left to right. For the
derivation tree in Figure 6.2, for example, this produces the program
y sin((x + z) t).
However, for efficiency reasons, in an actual implementation it is not convenient to extract the program represented by a derivation tree is this way.

6.2 Constraining Structures

55

This is because programs need to be executed in order to evaluate their

fitness, and this flat program representation often requires further transformations before execution. It is, therefore, common to directly convert a
derivation tree into a standard GP tree.
Grammar-based GP approaches can be extended by incorporating concepts from computational linguistics. For example, McKay and colleagues
used tree adjoining grammars (TAGs) (Joshi and Schabes, 1997) to design new genetic representations and operators that respect grammatical
constraints while allowing new types of structural modifications (Hoai and
McKay, 2004; Hoai et al., 2003; Hoai, McKay, Essam, and Hao, 2005).
Another major grammar-based approach is grammatical evolution (GE)
(ONeill and Ryan, 2003; Ryan, Collins, and ONeill, 1998). GE does not
use trees, instead it represents individuals as variable-length sequences of
integers (cf. Equation 6.4) which are interpreted in the context of a user
supplied grammar.
For each rule in the grammar, the set of alternatives on the right hand
side are numbered from 0 upwards. In the example grammar in Equation (6.3) above, the first rule only has one option on the right hand side; so
this would be numbered 0. The second rule has two options, which would
be numbered 0 and 1. The third rule has four options which would be numbered 0 to 3. Finally the fourth rule has three options numbered 0 to 2. To
create a program from a GE individual one uses the values in the individual
to choose which alternative to take in the production rules. For example,
suppose a GE individual is represented by the sequence
39, 7, 2, 83, 66, 92, 57, 80, 47, 94

(6.4)

then we start with 39 and the first syntax rule, tree. However tree has no
alternatives, so we move to 7 and rule E. Now E has two alternatives and
7 is used (via modulus) to chose between them. More of the translation
process is given in Figure 6.3.
In this example we did not need to use all the numbers in the sequence
to generate a complete program. Indeed the last integer, 94, was not used.
In general, extra genetic material is simply ignored. More problematic is
when a sequence is too short in the sense that the end of the sequence
is reached before the translation process is complete. There are a variety
of options in this case, including failure (assigning this individual the worst
possible fitness) and wrapping (continuing the translation process, moving
back to the front of the numeric sequence). Grammatical evolution has been
very successful and is widely used.

6.2.4

Constraints and Bias

One of the common arguments in favour of constraint systems like types

and grammars is that they limit the search space by restricting the kind

56

6 Modular, Grammatical and Developmental Tree-based GP

tree
h 39 mod 1 = 0, i.e., there is only one option i
E sin(E t)
h 7 mod 2 = 1, i.e., choose second option i
(E op E) sin(E t)
h 2 mod 2 = 0, i.e., take the first option i
(var op E) sin(E t)
h 83 mod 3 = 2, pick the third variable, i
(z op E) sin(E t)
h 66 mod 4 = 2, take the third operator i
(z E) sin(E t)

...
(z x) sin(z t)
Figure 6.3: Sample grammatical evolution derivation using the grammar in
Equation (6.3) and the integer sequence in Equation (6.4). The non-terminal
to be rewritten is underlined in each case.

of structures that can be constructed. While this is true, it can come at a

price.
An expressive type system typically requires more complex machinery to
support it. It often makes it more difficult to generate type-correct individuals in the initial population or during mutation and it is more difficult to
find crossover points that do not violate the type system. In an extreme case
like, constructive type theory (Thompson, 1991), the type system is so powerful that it can completely express the formal specification of the program.
Thus, any program/expression having this type is guaranteed to meet that
specification. In GP this would mean that all the members of the initial
population would need to be solutions to the problem, thus putting all the
problem solving burden on the initialisation phase and removing the need
for any evolution at all! Even without such extreme constraints, it has often
been found necessary to develop additional machinery in order to efficiently
generate an initial population that satisfies the constraints (Montana, 1995;
Ratle and Sebag, 2000; Schoenauer and Sebag, 2001; Yu, 2001).
As a rule, systems that focus on syntactic constraints (such as grammarbased systems) require less machinery than those that focus on semantic
constraints (such as type systems), since it is typically easier to satisfy the
syntactic constraints in a mechanistic fashion. For example, grammar-based

6.3 Developmental Genetic Programming

57

systems, such as grammatical evolution and the various TAG-based systems,

are typically simple to initialise, and mutation and crossover need to enforce
few, if any, constraints on the new child. The work (and the bias) in these
systems is much more in the choice of the grammar, and once it has been
designed, there is often little additional work required of the practitioner or
the GP system to enforce the implied constraints.
While a constraint system may limit the search space in valuable ways
(Ratle and Sebag, 2000) and can improve performance on interesting problems (Hoai, McKay, and Essam, 2006), there is no general guarantee that
constraint systems will make the evolutionary search process easier. There
is no broad assurance that a constraint will increase the density of solutions or (perhaps more importantly) approximate solutions.2 Also, while
there are cases where constraint systems smooth the search landscape (Hoai
et al., 2006), it is also possible for constraint systems to make the search
landscape more rugged by preventing genetic operations from creating intermediate forms on potentially valuable evolutionary paths. In the future, it
might be useful to extend solution density studies such as those summarised
in (Langdon and Poli, 2002) to the landscapes generated by constraint systems in order to better understand the impact of these constraints on the
underlying search spaces.
In summary, while types, grammars, and other constraint systems can
be powerful tools, all such systems carry biases. One therefore needs to be
careful to explore the biases introduced by the constraints and not simply
assume that they are beneficial.

6.3

Developmental Genetic Programming

By using appropriate terminals, functions and/or interpreters, GP can go

beyond the production of computer programs. In cellular encoding (Gruau,
1994; Gruau and Whitley, 1993; Gruau, 1994), programs are interpreted
as sequences of instructions which modify (grow) a simple initial structure
(embryo). Figure 6.4 shows part of the development of an electronic circuit.3
Once the program has finished, the quality of the structure it has produced
is measured and this is taken to be the fitness of the program.
Naturally, for cellular encoding to work the primitives of the language
must be able to grow structures appropriate to the problem domain. Typical
instructions involve the insertion and/or sizing of components, topological
2 By solution density we refer to the ratio between the number of acceptable solutions
in a program search space and the size of the search space itself. This is a rough assessment
of how hard a problem is, since it gives an indication of how long random search would
take to explore the program space before finding an acceptable solution.
3 The process is easier to explain with a movie. This can be downloaded from http:
//www.genetic-programming.com/gpdevelopment.html.

58

6 Modular, Grammatical and Developmental Tree-based GP

Figure 6.4: Screen shot of an animated gif showing the development

of the topology and the sizing of an electrical circuit (from http://www.
genetic-programming.com/gpdevelopment.html).The program is interpreted in parallel. Solid arrows link the active code to the parts of the
electronic circuit (lower half) that are being modified. The three-headed
arrow from S shows that three new components (Z4) have just been created
in series. Their types (e.g., capacitor, inductor or resistor) and values will
be determined by the three arguments of the S function.

modifications of the structure, etc. Cellular encoding GP has successfully

been used to evolve neural networks (Gruau, 1994; Gruau and Whitley,
1993; Gruau, 1994) and electronic circuits (Koza et al., 1999; Koza, Andre,
Bennett, and Keane, 1996a; Koza, Bennett, Andre, and Keane, 1996c), as
well as in numerous other domains. A related approach proposed by Hoang,
Essam, McKay, and Nguyen (2007) combines tree adjoining grammars (Section 6.2.3) with L-systems (Lindenmayer, 1968) to create a system where
each stage in the developmental process is a working program that respects
the grammatical constraints.
One of the advantages of indirect representations such as cellular encoding is that the standard GP operators can be used to evolve structures
(such as circuits) which may have nothing in common with standard GP
trees. In many of these systems, the structures being grown are also still
meaningful (and evaluable) at each point in their development. This allows
fitness evaluation. Another important advantage is that structures result-

6.4 Strongly Typed Autoconstructive GP with PushGP

59

ing from developmental processes often have some regularity, which other
methods obtain through the use of ADFs, constraints, types, etc. A disadvantage is that, with cellular encoding, individuals require an additional
genotype-to-phenotype decoding step. However, when the fitness function
involves complex calculations with many fitness cases, the relative cost of the
decoding step is often small compared with the rest of the fitness function.

6.4

Strongly Typed Autoconstructive GP

with PushGP

While types are often used to constrain evolution, Spectors PushGP (Klein
and Spector, 2007; Robinson and Spector, 2002; Spector, 2001; Spector,
Klein, and Keijzer, 2005a) is a move away from constraining evolution.
Essentially PushGP uses genetic programming to automatically create
programs written in the Push programming language. Push is a strongly
typed tree based language which does not enforce syntactic constraints.
Each of Pushs types has its own stack. In addition to stacks for integers, floats, Booleans and so on, there is a stack for objects of type program.
Using this code stack, Push naturally supports both recursion and program
modules (see Section 6.1.1) without human pre-specification. The code stack
allows an evolved program to push itself or fragments of itself onto the stack
for subsequent manipulation.
PushGP can use the code stack and other operations to allow programs to
construct their own crossover and other genetic operations and create their
own offspring. Programs are prevented from simply duplicating themselves
to deflect catastrophic loss of population diversity.

Chapter 7

Linear and Graph

Genetic Programming
Until now we have been talking about the evolution of programs expressed
as one or more trees which are evaluated by a suitable interpreter. This is
the original and most widespread type of GP, but there are other types of
GP where programs are represented in different ways. This chapter will look
at linear programs and graph-like (parallel) programs.

7.1

Linear Genetic Programming

In linear GP programs are linear sequences of instructions, such as the one

in Figure 7.1. The number of instructions can be fixed, meaning that every
program in the population has the same length, or variable, meaning that
different individuals can be of different sizes. In the following sections we
discuss reasons for using linear GP (Section 7.1.1). We then provide more
details on the different flavours of linear GP (Section 7.1.2). Finally, we
describe briefly the main genetic operations for linear GP (Section 7.1.3).

7.1.1

Motivations

There are two different reasons for trying linear GP. Firstly, almost all
computer architectures represent computer programs in a linear fashion with

....

Instruction 1 Instruction 2

Instruction N

Figure 7.1: Typical linear GP representation for programs.

61

62

7 Linear and Graph Genetic Programming

Output
R0..R7

Arg 1
R0..R7

Opcode
+*/

Arg 2
0...127
or
R0..R7

Figure 7.2: Format of a linear GP engine instruction. R0 to R7 refer to

CPUs registers.
neighbouring instructions being normally executed in consecutive time steps
(albeit control structures, jumps and loops may change the execution order).
So, why not evolve linear programs? This led Banzhaf (1993), Perkis (1994)
as well as Openshaw and Turton (1994) to try linear GP.
Secondly, computers do not naturally run tree-shaped programs, so interpreters or compilers have to be used as part of tree-based GP. On the
contrary, by evolving the binary bit patterns actually obeyed by the computer, linear GP can avoid the use of this computationally expensive machinery and GP can run several orders of magnitude faster. This desire for
speed drove Nordin (1994), Nordin et al. (1999), Crepeau (1995) and Eklund
(2002).

7.1.2

Linear GP Representations

As discussed in Section 2.1, it is possible to use a linear representation in

tree-based GP. When doing so, however, the linear structures are simply
flattened representations of the trees. Thus, in the linear structure one can
still identify the root node, its children, and the rest of the tree structure.
In such a system, instructions typically communicate via their arguments.
The semantics of linear GP are quite different, however. In linear GP, instructions typically read their input(s) from one or more registers or memory
locations and store the results of their calculations in a register. For example, they might take the form shown in Figure 7.2. This means instructions
in linear GP all have equivalent roles and communicate only via registers
or memory. In linear GP there is no equivalent of the distinction between
functions and terminals inherent in tree-based GP. Also, in the absence of
loops or branches, the position of the instructions determines the order of
their execution. Typically, this not the case for the structures representing
trees.1
The instructions in linear GP may or may not represent executable machine code. That is, there are essentially two flavour of linear GP: machine
code GP, where each instruction is directly executable by the CPU, and
1 Typically, in tree-based-GP the nodes are visited (but not executed) from left to right
in depth-first order. Primitives are only executed, however, when their arguments have
been evaluated. So, the root node is the first node visited but the last executed.

7.1 Linear Genetic Programming

63

interpreted linear GP, where each instruction is executable by some higherlevel virtual machine (typically written in an efficient language such as C
or C++). When the instructions are actual machine code, then the order
of the elements of the representation shown in Figure 7.2 is determined by
the particular computer architecture used, and the corresponding data must
be packed into bit fields of appropriate sizes. The overhead of packing and
unpacking data can be avoided, however, when one is using virtual machine
instructions since then the designer of a GP system has complete freedom
as to how the virtual machine will interpret its instructions.
If the goal is execution speed, then the evolved code should be machine
code for a real computer rather than some higher level language or virtualmachine code. This is why Nordin (1994) started by evolving machine code
for SUN computers and Crepeau (1995) targeted the Z80. The linear GP
of Leung, Lee, and Cheang (2002) was designed for novel hardware, but much
of the GP development had to be run in simulation whilst the hardware itself
was under development.
The Sun SPARC has a simple 32-bit RISC architecture which eases
designing genetic operations which manipulate its machine code. Nordin
(1997) wrapped each machine code GP individual (which was a sequence of
machine instructions) inside a C function. Each of the GP programs inputs
was copied from one of the C functions arguments into one of the machine
registers. As well as the registers used for inputs,2 a small number (e.g.,
24) of other registers are used for scratch memory to store partial results of
intermediate calculations. Finally, the GP simply leaves its answer in one of
the registers. The external framework uses this as the C functions return
value.
Since Unix was ported onto the x86, Intels complex instruction set,
which was already standard with Windows-based PCs, has had almost complete dominance. Seeing this, Nordin ported his Sun RISC linear GP system
onto Intels CISC. Various changes were made to the genetic operations
which ensure that the initial random programs are made only of legal Intel machine code and that mutation operations, which act inside the x86s
32-bit word, respect the x86s complex sub-fields. Since the x86 has instructions of different lengths, special care has to be taken when altering them.
Typically, several short instructions are packed into each 4-byte word. If
there are any bytes left over, they are filled with no-operation codes. In
this way, best use is made of the available space, without instructions crossing 32-bit boundaries. Nordins work led to Discipulus (Foster, 2001),
which has been used in applications ranging from bioinformatics (Vukusic,
Grellscheid, and Wiehe, 2007) to robotics (Langdon and Nordin, 2001) and
2 Anyone using a register-based GP (linear or tree-based) should consider writeprotecting the input registers to prevent the inputs from being overwritten. Otherwise
evolved programs (especially in the early generations) are prone to writing over their
inputs before theyve had a chance to use them in any constructive way.

64

7 Linear and Graph Genetic Programming

bomb disposal (Deschaine, Hoover, Skibinski, Patel, Francone, Nordin, and

Ades, 2002).
Note that execution speed is not the only reason for using linear GP. Although interpreted linear programs are slower than machine-code programs,
an interpreted linear GP system can be more efficient than an interpreted
tree-based systems. Also, a simple linear structure lends itself to rapid analysis. Brameier and Banzhaf (2001) showed a linear program can be easily
scanned and any dead code it contains can be removed. In some ways the
search space of linear GP is also easier to analyse than that of trees (Langdon, 1999b, 2002a,b, 2003a; Langdon and Banzhaf, 2005). For example,
Langdon (2006); Langdon and Poli (2006,?, 2008) have used (in simulation)
two simple architectures, called T7 and T8 ), for several large scale experiments and for the mathematical analysis of Turing complete GP. For these
reasons, it makes sense to consider linear virtual-machine code GP even
when using languages like Java that are typically run on virtual machines;
one can in fact use a virtual machine (like (Leung et al., 2002)) to interpret the evolved byte code (Harvey, Foster, and Frincke, 1999; Lukschandl,
Borgvall, Nohle, Nordahl, and Nordin, 2000).

7.1.3

Linear GP Operators

The typical crossover and mutation operators for linear GP ignore the details
of the machine code of the computer being used. For example, crossover may
choose randomly two crossover points in each parent and swaps the code
between them. Since the crossed over fragments are typically of different
lengths, such a crossover may change the programs lengths, cf. Figure 7.3.
Since computer machine code is organised into 32- or 64-bit words, the
crossover points occur only at the boundaries between words. Therefore,
a whole number of words, containing a whole number of instructions are
typically swapped over. Similarly, mutation operations normally respect
word boundaries and generate legal machine code. However, linear GP lends
itself to a variety of other genetic operations. For example, Figure 7.4 shows
homologous crossover. Many other crossover and mutation operations are
possible (Langdon and Banzhaf, 2005).
In a compiling genetic programming system (Banzhaf, Francone, and
Nordin, 1996) the mutation operator acts on machine code instructions
and is constrained to ensure that only instructions in the function set are
generated and that the register and constant values are within predefined
ranges allowed in the experimental set up. On some classification problems Banzhaf et al. (1996) reported that performance was best when using
crossover and mutation in equal proportions. They suggested that this was
due to the GP population creating introns (blocks of code that does not
affect fitness) in response to the crossover operator, and that these were subsequently converted into useful genetic material by their mutation operator.

7.2 Graph-Based Genetic Programming

65

Parent 1
Parent 2

Offspring
Figure 7.3: Typical linear GP crossover. Two instructions are randomly
chosen in each parent (top two genomes) as cut points. The code fragment
excised from the first parent is then replaced with the code fragment excised
from the second to generate the child (lower chromosome).
Parent 1
Parent 2

Offspring 1
Offspring 2
Figure 7.4: Discipuluss homologous crossover (Foster, 2001; Francone
et al., 1999; Nordin et al., 1999). Crossover is performed on two parents (top
two programs) to yield two offspring (bottom). The two crossover points are
the same in both parents, so the exised code does not change its position
relative to the start of the program (left edge), and the child programs have
the same lengths as their parents. Homologous crossover is often combined
with a small amount of normal two point crossover (Figure 7.3) to introduce
length changes into the GP population.

7.2

Graph-Based Genetic Programming

Trees are special types of graphs. So it is natural to ask what would happen
if one extended GP so as to be able to evolve graph-like programs. Starting
from the mid 1990s, researchers have proposed several extensions of GP that
do just that, albeit in different ways.

7.2.1

Parallel Distributed GP

Poli (1996a, 1999a) proposed parallel distributed GP (PDGP), a form of GP

that is suitable for the evolution of highly parallel programs which effectively reuse partial results. Programs are represented in PDGP as graphs

66

7 Linear and Graph Genetic Programming

max

max

y
x
(a)

3
y

3
y
(b)

Figure 7.5: A sample tree where the same subtree is used twice (a) and
the corresponding graph-based representation of the same program (b). The
graph representation may be more efficient since it makes it possible to avoid
the repeated evaluation of the same subtree.
with nodes representing functions and terminals. Edges represent both control flow and data flow. The possible efficiency gains obtained by a graph
representation are illustrated in Figure 7.5.
In the simplest form of PDGP edges are directed and unlabelled, in
which case PDGP is a generalisation of standard GP. However, more complex representations can be used, which allow the evolution of: programs,
including standard tree-like programs, logic networks, neural networks, recurrent transition networks and finite state automata. This can be achieved
by extending the representation by associating labels with the edges of the
program graph. In addition to the function and terminal sets, this form of
PDGP requires the definition of a link set. The labels on the links depend
on what is to be evolved. For example, in neural networks, the link labels
are numerical constants for the neural network weights. In a finite state automaton, the edges are labelled with the input symbols that determine the
FSAs state transitions. It is even possible for the labels to be automatically
defined edges, which play a role similar to ADFs (Section 6.1.1) by invoking
other PDGP graphs.
In PDGP, programs are manipulated by special crossover and mutation
operators which guarantee the syntactic correctness of the offspring. Each
node occupies a position in a regular grid. The genetic operators act by
moving, copying or randomly generating sub-regions of the grid. For this
reason PDGP search operators are very efficient.
PDGP programs can be executed according to different policies depend-

7.2 Graph-Based Genetic Programming

67

ing on whether instructions with side effects are used or not. If there are
no side effects, running a PDGP program can be seen as a propagation of
the input values from the bottom to the top of the programs graph (as in
a feed-forward artificial neural network or data flow parallel computer).

7.2.2

Parallel Algorithm Discovery and Orchestration

In a system called parallel algorithm discovery and orchestration (PADO),

Teller (1996) used a combination of GP and linear discrimination to obtain
parallel classification programs for signals and images.
PADO programs include three parts: a main loop, some ADFs and an
indexed memory. The main loop is repeatedly executed for a fixed amount
of time. When the time is up, PADO programs are forced to halt by some
external control structure. The output of a program is the weighted average
of the outputs produced at each iteration of the loop. The weights are
proportional to the iteration count, so that more recent outputs count more.
The main loop and the ADFs in PADO are structured as arbitrary directed graphs of nodes. Each node can have multiple outgoing arcs that
indicate possible flows of control. Each node has two main parts: an action and a branch-decision. Each program has an argument stack and all
PADO actions pop their inputs from this argument stack and push their result back onto the argument stack. The actions are drawn from a primitive
set including the standard algebraic operations, minimum, maximum, negation, read from indexed memory, write to indexed memory, deterministic
and non-deterministic branching instructions, and primitives related to the
task of classifying images. The evaluation of PADO programs starts from a
designated node. After the execution of each node, control is passed to the
node chosen by the branch-decision function of the current node.

7.2.3

Cartesian GP

In Millers Cartesian GP (Miller, 1999; Miller and Smith, 2006), programs

are represented by linear chromosomes containing integers. These are divided into groups of three or four. Each group corresponds to a position in
a 2-D array. One integer in each group defines the primitive (e.g., an AND
gate) at that location in the array. Other integers in the group define the
locations (coordinates) in the genome from which the inputs for that primitive should be drawn. Each primitive does not itself define where its output
is used; this is done by later primitives. A primitives output may be used
more than once, or indeed not used at all, depending on the way in which
the other functions inputs are specified. Thus, Cartesian GPs chromosomes represent graph-like programs, which is very similar to PDGP. The
main difference between the two systems is that Cartesian GP operators act
at the level of the linear chromosome, while in PDGP they act directly on

68

7 Linear and Graph Genetic Programming

the graph. Also, traditionally Cartesian GP has always used mutation as

its main search operation, while PDGP used both crossover and mutation.
However, recently a new crossover has been proposed for Cartesian GP that
provides faster convergence (Clegg, Walker, and Miller, 2007).

7.2.4

Evolving Parallel Programs using Indirect

Encodings

The graph-based systems discussed above use representations which directly

encode parallel programs. However, it is also possible to use non-graphbased GP to evolve parallel programs. For example, Bennett (1996) used a
parallel virtual machine in which several standard tree-like programs (called
agents) would have their nodes executed in parallel. He included a twostage mechanism which simulated parallelism of sensing actions and simple
conflict resolution (prioritisation) for actions with side effects. Andre, Bennett, and Koza (1996) used GP to discover rules for cellular automata, a
highly parallel computational architecture, which could solve large majorityclassification problems. In conjunction with an interpreter implementing a
parallel virtual machine, GP can also be used to translate sequential programs into parallel ones (Walsh and Ryan, 1996), or to develop parallel
programs.

Chapter 8

Probabilistic Genetic
Programming
Genetic programming typically uses an evolutionary algorithm as its main
search engine. However, this is not the only option. The use of simulated
annealing and hill climbing to search the space of computer programs was
mentioned in Section 5.4. This chapter considers recent work where the exploration is performed by population-based search algorithms which adapt
and sample probability distributions instead of using traditional genetic operators.
Sampling from a probability distribution means generating random values whose statistical properties match those of the given distribution. For
example, if one sampled a univariate Gaussian distribution, one would expect the resulting values to tend to have mean and standard deviation similar to the mean and standard deviation of the Gaussian. The notion of
sampling can be extended to much more complex distributions involving
multiple variables. Furthermore, discrete as well as continuous variables are
possible.

8.1

Estimation of Distribution Algorithms

Estimation of distribution algorithms (EDAs) are powerful population-based

searchers where the variation operations traditionally implemented via
crossover and mutation in EAs are replaced by the process of random sampling from a probability distribution. The distribution is modified generation after generation, using information obtained from the fitter individuals
in the population. The objective of these changes in the distribution is to
increase the probability of generating individuals with high fitness.
69

70

8 Probabilistic Genetic Programming

Different EDAs use different models for the probability distribution that
controls the sampling (see (Larra
naga, 2002; Larra
naga and Lozano, 2002)
for more information). For example, population-based incremental learning
(PBIL) (Baluja and Caruana, 1995) and the uniform multivariate distribution algorithm (UMDA) (M
uhlenbein and Mahnig, 1999a,b) assume that
each variable is independent of the other variables. Consequently, these algorithms need to store and adjust only a linear array of probabilities, one for
each variable. This works well for problems with weak interactions between
variables. Since no relationship between the variables is stored or learned,
however, PBIL and UMDA may have difficulties solving problems where the
interactions between variables are significant.
Naturally, higher order models are possible. For example, the MIMIC
algorithm of de Bonet, Isbell, and Viola (1997) uses second-order statistics. It is also possible to use flexible models where interactions of different orders are captured. The Bayesian optimisation algorithm (BOA)
(Pelikan, Goldberg, and Cant
u-Paz, 1999) uses baysian networks to represent generic sampling distributions, while the extended compact genetic
algorithm (eCGA) (Harik, 1999) clusters genes into groups where the genes
in each group are assumed to be linked but groups are considered independent. The sampling distribution is then taken to be the product of the
distributions modelling the groups.
EDAs have been very successful. However, they are often unable to represent both the overall structure of the distribution and its local details,
typically being more successful at the former. This is because EDAs represent the sampling distribution using models with an, inevitably, limited
number of degrees of freedom. For example, suppose the optimal sampling
distribution has multiple peaks, corresponding to different local optima, separated by large unfit areas. Then, an EDA can either decide to represent
only one peak, or to represent all of them together with the unfit areas. If
the EDA chooses the wrong local peak this may lead to it getting stuck and
not finding the global optimum. Conversely if it takes a wider view, this
leads to wasting many trials sampling irrelevant poor solutions.
Consider, for example, a scenario where there are five binary variables,
x1 , x2 , x3 , x4 and x5 , and two promising regions: one near the string of all
zeros, i.e., (x1 , x2 , x3 , x4 , x5 ) = (0, 0, 0, 0, 0), and the other near the string
of all ones, i.e., (x1 , x2 , x3 , x4 , x5 ) = (1, 1, 1, 1, 1). One option for a (simple)
EDA is to focus on one of the two regions, e.g., setting the variables xi
to 0 with high probability (say, 90%). This, however, fails to explore the
other region, and risks missing the global optimum. The other option is to
maintain both regions as possibilities by setting all the probabilities to 50%,
i.e., each of the variables xi is as likely to be 0 as 1. These probabilities will
generate samples in both of the promising regions. For example, the strings
(0, 0, 0, 0, 0) and (1, 1, 1, 1, 1) will each be generated with a 3.125% proba-

8.2 Pure EDA GP

71

bility. Also, simple calculations show that 31.25% of individuals generated

by this distribution will be at Hamming distance 1 from either (0, 0, 0, 0, 0)
or (1, 1, 1, 1, 1).1 So, both optimal regions are sampled reasonably often.
However, it is clear that the majority (62.5%) of samples will be allocated
to less promising regions, where the Hamming distance will be 2 or 3 from
both (0, 0, 0, 0, 0) and (1, 1, 1, 1, 1). This is a significant concern, which is
why recently EDAs have often been used in combination with local search
(e.g., see (Zhang, Sun, and Tsang, 2005)).
There have been several applications of probabilistic model-based evolution (EDA-style) in the areas of tree-based and linear GP. We review them
in the rest of this chapter.

8.2

Pure EDA GP

The first EDA-type GP system was effectively an extension of PBIL to trees

called probabilistic incremental program evolution (PIPE) (Salustowicz and
Schmidhuber, 1997; Salustowicz, Wiering, and Schmidhuber, 1998; Salustowicz and Schmidhuber, 1999). In PIPE, the population is replaced by a
hierarchy of probability tables organised into a tree (such as the one in Figure 8.1). Each table represents the probability that a particular primitive
will be chosen at that specific location in a newly generated program tree.
At each generation a population of programs is created based on the current
tree of probability tables. The generation of a program begins by choosing a
root node based on the probabilities in the root table, and then continuing
down the hierarchy of probability tables until all branches of the tree are
complete (i.e., a terminal has been chosen on each branch). The fitness of
these new programs is computed, and the probability hierarchy is updated
on the basis of these fitnesses, so as to make the generation of above-average
fitness programs more likely in the next generation.
A positive feature of PIPE is that the probability of choosing a particular
primitive can vary with its depth (and, more generally, position) in the tree.
This makes it possible, for example, for terminals to become increasingly
probable as a nodes depth increases. A limitation of PIPE, however, is that
the primitives forming a tree are chosen independently from each other,2 so it
is impossible for PIPE to capture dependencies among primitives. Another
limitation is that the maximum size of the generated trees is constrained
by the size of the tree of probability tables. Ondas, Pelikan, and Sastry
(2005) compared the performance of PIPE and standard tree-based GP on
1 The

Hamming distance between two strings (whether binary or not) is the number
of positions where the two strings differ.
2 There is a weak form of dependency, in that there can be a primitive in a particular
position only if the primitive just above it is a function. The choice of this parent primitive
does not, however, influence the choice of the child primitive.

72

8 Probabilistic Genetic Programming

a small set of artificial problems, including a GP version of one-max3 and a

GP version of the fully deceptive trap function4 . Results suggest that PIPE
and standard GP have similar scaling properties, but that standard subtree
crossover inherently links neighbouring nodes whereas PIPE does not.
Sastry and Goldberg (2003) proposed an algorithm called extended compact GP (eCGP) which effectively extends the eCGA algorithm (Harik,
1999) to trees. Like PIPE, eCGA assumes that all trees will fit within a
fixed maximal tree. It partitions the nodes in this maximal tree into groups.
The nodes in a group are assumed to be linked and their co-occurrence is
modelled by a full joint distribution table. As with eCGA, the probability
of generating a particular tree is given by the product of the probabilities of
generating each group of nodes using the groups joint distributions. An advantage of this system is that, unlike PIPE, it captures dependencies among
primitives. However, to the best of our knowledge this system has only been
tested on the two artificial problems used by Ondas et al. (2005) to compare
PIPE and GP. Consequently its behaviour on more typical GP problems is
unknown.
Yanai and Iba (2003) proposed an EDA called estimation of distribution
programming (EDP) which, in principle, can capture complex dependencies
between a node in a program tree and the nodes directly above it or to its
left.5 As with eCGP and PIPE, programs are tree-like and are assumed
to always fit within an ideal maximal full tree. A conditional probability
table is necessary for each node in such a tree to capture the dependencies.
To keep the size of data structures manageable, only pairwise dependencies
between each node and its parent were stored and used. EDP was tested on
both the Max problem (Gathercole and Ross, 1995) and the 6-multiplexer.
A later hybrid algorithm combining EDP and GP was proposed (Yanai and
Iba, 2004) which showed promise when tested on three symbolic regression
problems.
An EDA based on a hierarchical BOA was used as the main mechanism
to generate new individuals in meta-optimising semantic evolutionary search
(MOSES) (Looks, 2007). This combined multiple strategies and used semantics to restrict and direct the search. BOA was also used to evolve programs
in (Looks, Goertzel, and Pennachin, 2005) using a specialised representation
for trees.
3 One-max is a simple GA test problem where the goal is to maximise the number of
1s in a binary string.
4 Trap functions are fitness functions that have a gradual slope leading to a sub-optimal
local maxima, and a steep valley between that local maxima and the global optima. They
therefore tend to trap populations on the local maxima
5 In the general case a node can depend on the choices of any of the nodes that have
already been chosen. Since the tree is constructed in a depth-first, left-to-right fashion,
it can depend on any nodes that are its direct ancestors, or any nodes that are to its left
in the tree. In practice, however, EDP only tracked the conditional probability of a node
on its parent.

8.2 Pure EDA GP

73

+
*

/
x

+
*

/
x

+
*

/
x

0.1
0.2
0.1
0.0
0.4
0.2

0.1
0.2
0.1
0.2
0.3
0.1

+
*

0.2
0.3

0.1
0.1
0.2
0.1

+
*

/
x

/
x

R
0.0
0.0
0.3
0.1
0.1
0.5

+
*

/
x

0.1
0.1
0.2
0.1
0.2
0.3

0.1
0.1
0.1
0.4
0.3
0.0

+
*

/
x

0.3
0.1
0.0
0.0
0.3
0.3

Figure 8.1: Example of probability tree used for the generation of programs
in PIPE. New program trees are created starting from the root node at the
top and moving through the hierarchy. Each node in an offspring tree is
selected from the left hand side of the corresponding table with probability
given by the right hand side. Each branch of the tree continues to expand
until either the tree of probability tables is exhausted or a leaf (e.g., R) is
selected.

74

8 Probabilistic Genetic Programming

Recently an EDA-based system called N-gram GP (Poli and McPhee,

2008a) has been proposed that allows the evolution of linear GP programs.
To some extent, N-gram GP overcomes the common difficulties EDAs have
in performing local search when using a centralised population model. The
N-gram GP system is able to capture both the local and the global features
of the optimal sampling distribution, albeit at the cost of imposing certain
other constraints. This makes it possible, for example, for the search to focus
on the neighbourhood of a small number of individuals without the need to
choose among them. Tests on polynomial symbolic regression problems and
the lawnmower problem were very encouraging.

8.3

Mixing Grammars and Probabilities

A variety of other systems have been proposed which combine the use of
grammars and probabilities. We mention only a few here; a more extended
review of these is available in (Shan, McKay, Essam, and Abbass, 2006).
Ratle and Sebag (2001) used a stochastic context-free grammar to generate program trees. The probability of applying each rewrite rule was
adapted using a standard EDA approach so as to increase the likelihood of
using successful rules. The system could also be run in a mode where rule
probabilities depended upon the depth of the non-terminal symbol to which
a rewrite rule was applied, thereby providing a higher degree of flexibility.
The approach taken in program evolution with explicit learning (PEEL)
(Shan, McKay, Abbass, and Essam, 2003) was slightly more general. PEEL
used a probabilistic L-system where rewrite rules were both depth- and
location-dependent. The probabilities with which rules were applied were
adapted by an ant colony optimisation (ACO) algorithm (Dorigo and
St
utzle, 2004). Another feature of PEEL was that the L-systems rules
could be automatically refined via splitting and specialisation.
Other programming systems based on probabilistic grammars which are
optimised via ant systems include ant-TAG (Abbass, Hoai, and McKay,
2002; Shan, Abbass, McKay, and Essam, 2002), which uses a tree-adjunct
grammar as its main representation, and generalised ant programming
(GAP) (Keber and Schuster, 2002), which is based on a context-free grammar. Other systems which learn and use probabilistic grammars include
grammar model based program evolution (GMPE) (Shan, McKay, Baxter,
Abbass, Essam, and Hoai, 2004), the system described in (Bosman and de
Jong, 2004a,b) and Baysian automatic programming (BAP) (Regolin and
Pozo, 2005).

Chapter 9

Multi-objective
Genetic Programming
The area of multi-objective GP (MO GP) has been very active in the last
decade. In a multi-objective optimisation (MOO) problem, one optimises
with respect to multiple goals or fitness functions f1 , f2 , .... The task of a
MOO algorithm is to find solutions that are optimal, or at least acceptable,
according to all the criteria simultaneously.
In most cases changing an algorithm from single-objective to multiobjective requires some alteration in the way selection is performed. This is
how many MO GP systems deal with multiple objectives. However, there
are other options. We review the main techniques in the following sections.
The complexity of evolved solutions is one of the most difficult things
to control in evolutionary systems such as GP, where the size and shape of
the evolved solutions is under the control of evolution. In some cases, for
example, the size of the evolved solutions may grow rapidly, as if evolution
was actively promoting it, without any clear benefit in terms of fitness. We
will provide a detailed discussion of this phenomenon, which is know as bloat,
and a variety of counter measures for it in Section 11.3. However, in this
chapter we will review work where the size of evolved solutions has been
used as an additional objective in multi-objective GP systems. Of course,
we will also describe work where other objectives were used.

9.1

Combining Multiple Objectives into a

Scalar Fitness Function

When given multiple fitness functions, it is natural to think of combining

them in some way so as to produce an aggregate scalar fitness function. For
75

76

9 Multi-objective Genetic Programming

P
example, one could use a linear combination of the form f = i wi fi , where
the parameters w1 , w2 , . . . are suitable constants. A MOO problem can then
be solved by using any single-objective optimisation technique with f as a
fitness function. This method has been used frequently in GP to control
bloat. By combining program fitness and program size to form a parsimonious fitness function one can evolve solutions that satisfy both objectives
(see Koza (1992); Zhang and M
uhlenbein (1993, 1995); Zhang, Ohm, and
M
uhlenbein (1997) and Section 11.3.2).
A semi-linear aggregation of fitness and speed was used in (Langdon
and Poli, 1998b) to improve the performance of GP on the Santa Fe Trail
Ant problem. There, a threshold was used to limit the impact of speed to
avoid providing an excessive bias towards ants that were fast but could not
complete the trail.
A fitness measure which linearly combines two related objectives, the
sum of squared errors and the number of hits (a hit is a fitness case in which
the error falls below a pre-defined threshold), was used in (Langdon, Barrett,
and Buxton, 2003) to predict biochemical interactions in drug discovery.
Zhang and Bhowan (2004) used a MO GP approach for object detection.
Their fitness function was a linear combination of the detection rate (the
percentage of small objects correctly reported), the false alarm rate (the
percentage of non-objects incorrectly reported as objects), and the false
alarm area (the number of false alarm pixels which were not object centres
but were incorrectly reported as object centres).
OReilly and Hemberg (2007) used six objectives for the evolution of
L-systems which developed into 3-D surfaces in response to a simulated
environment. The objectives included the size of the surface, its smoothness,
its symmetry, its undulation, the degree of subdivision of the surface, and
the softness of its boundaries.
(Koza, Jones, Keane, and Streeter, 2004) used 16 different objectives
in the process of designing analogue electrical circuits. In the case of an
amplifier circuit these included: the 10dB initial gain, the supply current, the
offset voltage, the gain ratio, the output swing, the variable load resistance
signal output, etc. These objectives were combined in a complex heuristic
way into a scalar fitness measure. In particular, objectives were divided
into groups and many objectives were treated as penalties that were applied
to the main fitness components only if they are outside certain acceptable
tolerances.

9.2

Keeping the Objectives Separate

Since selection does not depend upon how the members of the population
are represented, the MOO techniques developed for other evolutionary algorithms can be easily adapted to GP.

9.2 Keeping the Objectives Separate

77

Pareto Front
y

Non dominated
points

1
2
B

Figure 9.1: Two-dimensional example of Pareto optimality and the Pareto

front, where the goal is to maximise along both the x and y axes. Solutions
A and B do not dominate each other. However, solution B is dominated by
solution 2. (Adapted from (Langdon, 1998).)

The main idea in MOO is the notion of Pareto dominance. Given a set
of objectives, a solution is said to Pareto dominate another if the first is not
inferior to the second in all objectives, and, additionally, there is at least one
objective where it is better. This notion can lead to a partial order, where
there is no longer a strict linear ordering of solutions. In Figure 9.1, for
example, individual A dominates (is better than) individual B along the y
axis, but B dominates A along the x axis. Thus there is no simple ordering
between then. The individual marked 2, however dominates B on both
axes and would thus be considered strictly better than B.
In this case the goal of the search algorithm becomes the identification of
a set of solutions which are non-dominated by any others. Ideally, one would
want to find the Pareto front, i.e., the set of all non-dominated solutions in
the search space. However, this is often unrealistic, as the size of the Pareto
front is often limited only by the precision of the problem representation. If
x and y in Figure 9.1 are real-valued, for example, and the Pareto front is
a continuous curve, then it contains an infinite number of points, making a
complete enumeration impossible.

9.2.1

Multi-objective Bloat and Complexity Control

Rodriguez-Vazquez, Fonseca, and Fleming (1997) performed non-linear system identification using a MO GP system, where individuals were selected
based on the Pareto dominance idea. The two objectives used were fitness
and model complexity. In each generation individuals were ranked based on
how many other individuals dominated them, and fitness was based on their
rank. To better cover the Pareto front, niching via fitness sharing (Goldberg, 1989) was also performed. Preference information was also included

78

9 Multi-objective Genetic Programming

to focus the selection procedure towards specific regions of the Pareto front.
Hinchliffe, Willis, and Tham (1998) applied similar ideas to evolve parsimonious and accurate models of chemical processes using MO GP. Langdon
and Nordin (2000) applied Pareto tournaments to obtain compact solutions
in programmatic image compression, two machine learning benchmark problems and a consumer profiling task. Nicolotti, Gillet, Fleming, and Green
(2002) used multi-objective GP to evolve quantitative structureactivity relationship models in chemistry; objectives included model fitting, the total
number of terms and the occurrence of non-linear terms.
Ekart and Nemeth (2001) tried to control bloat using a variant of Pareto
tournament selection where an individual is selected if it is not dominated
by a set of randomly chosen individuals. If the test fails, another individual
is picked from the population, until one that is non-dominated is found.
In order to prevent very small individuals from taking over the population
in the early generations of runs, the Pareto criterion was modified so as
to consider as non-dominated solutions also those that were only slightly
bigger, provided their fitness was not worse.
Bleuler, Brack, Thiele, and Zitzler (2001) suggested using the well-known
multi-objective optimiser SPEA2 (Zitzler, Laumanns, and Thiele, 2001) to
reduce bloat. de Jong, Watson, and Pollack (2001) and de Jong and Pollack
(2003) proposed using a multi-objective approach to promote diversity and
reduce bloat, stressing that without diversity enhancement (given by modern
MOO methods) searches can easily converge to solutions that are too small
to solve a problem. Tests with even parity and other problems were very
encouraging. Badran and Rockett (2007) argued in favour of using mutation
to prevent the population from collapsing onto single-node individuals when
using a multi-objective GP.
As well as directly fighting bloat, MO GP can also be used to simplify
solution trees. After GP has found a suitable (but large) model, for example,
one can continue the evolutionary process, changing the fitness function to
include a second objective that the model be as small as possible (Langdon,
1998). GP can then trim the trees while ensuring that the simplified program
still fits the training data.

9.2.2

Other Objectives

Although much of the use of MOO techniques in GP has been aimed at

controlling bloat, there are also genuinely MOO applications.
For example, Langdon (1998) made extensive use of Pareto dominance
ranking to evolve different types of data structures. Up to six different
criteria were used to indicate to what degree an evolved data structure met
the requirements of the target data structure. The criteria were used in
Pareto-type tournament selection, where, unlike in other systems, a second

9.2 Keeping the Objectives Separate

79

round of comparisons with the rest of the population was used as a tie
breaker. The method successfully evolved queues, lists, and circular lists.
Langdon and Poli (1998b) used Pareto selection with two objectives,
fitness and speed, to improve the performance of GP on the Santa Fe Trail
Ant problem. Ross and Zhu (2004) used MO GP with different variants of
Pareto selection to evolve 2-D textures. The objectives were feature tests
that were used during fitness evaluation to rate how closely a candidate
texture matched visual characteristics of a target texture image. Dimopoulos
(2005) used MO GP to identify the Pareto set for a cell-formation problem
related to the design of a cellular manufacturing production system. The
objectives included the minimisation of total intercell part movement, and
the minimisation of within-cell load variation.
Rossi, Liberali, and Tettamanzi (2001) used MO GP in electronic design
automation to evolve VHDL code. The objectives used were the suitability
of the filter transfer function and the transition activity of digital blocks.
Cordon, Herrera-Viedma, and Luque (2002) used Pareto-dominance-based
GP to learn Boolean queries in information retrieval systems. They used two
objectives: precision (the ratio between the relevant documents retrieved in
response to a query and the total number of documents retrieved) and recall
(the ratio between the relevant documents retrieved and the total number
of documents relevant to the query in the database).
Barlow (2004) used a GP extension of the well-known NSGA-II MOO
algorithm (Deb, Agrawal, Pratap, and Meyarivan, 2000) for the evolution of
autonomous navigation controllers for unmanned aerial vehicles. Their task
was locating radar stations, and all work was done using simulators. Four
objectives were used: the normalised distance from the emitter, the circling
distance from the emitter, the stability of the flight, and the efficiency of
the flight.
Araujo (2006) used MO GP for the joint solution of the tasks of statistical
parsing and tagging of natural language. Their results suggest that solving
these tasks jointly led to better results than approaching them individually.
Han, Zhou, and Wang (2006) used a MO GP approach for the identification of chaotic systems where the objectives included chaotic invariants
obtained by chaotic time series analysis as well, as the complexity and performance of the models.
Khan (2006) used MO GP to evolve digital watermarking programs. The
objectives were robustness in the decoding stage, and imperceptibility by the
human visual system. Khan and Mirza (2007) added a third objective aimed
at increasing the strength of the watermark in relation to attacks.
Kotanchek, Smits, and Vladislavleva (2006) compared different flavours
of Pareto-based GP systems in the symbolic regression of industrial data.
Weise and Geihs (2006) used MO GP to evolve protocols in sensor networks.
The goal was to identify one node on a network to act as a communication

80

9 Multi-objective Genetic Programming

relay. The following objectives were used: the number of nodes that know
the designated node after a given amount of time, the size of the protocol
code, its memory requirements, and a transmission count.
Agapitos, Togelius, and Lucas (2007) used MO GP to encourage the
effective use of state variables in the evolution of controllers for toy car
racing. Three different objectives were used: the ratio of the number of
variables used within a program to the number of variables offered for use by
the primitive language, the ratio of the number of variables being set within
the program to the number of variables being accessed, and the average
positional distance between memory setting instructions and corresponding
memory reading instructions.
When two or three objectives need to be simultaneously optimised, the
Pareto front produced by an algorithm is often easy to visualise. When
more than three objectives are optimised, however, it becomes difficult to
directly visualise the set of non-dominated solutions. Valdes and Barton
(2006) proposed using GP to identify similarity mappings between highdimensional Pareto fronts and 3-D space, and then use virtual reality to
visualise the result.

9.2.3

Non-Pareto Criteria

Pareto dominance is not the only way to deal with multiple objectives without aggregating them into a scalar fitness function.
Schmiedle, Drechsler, Grosse, and Drechsler (2001) compared GP with
four different MOO selection methods on the identification of binary decision diagrams. Linear weighting of the objectives was compared against: a)
Pareto dominance; b) a weaker form of Pareto dominance where a solution
is preferred to another if the number of objectives where the first is superior
to the second is bigger than the number of objectives where the opposite is
true; c) lexicographic ordering (where objectives are ordered based on the
users preference); and d) a new method based on priorities. The lexicographic parsimony pressure method proposed in (Luke and Panait, 2002;
Ryan, 1994) is in fact a form of MOO with lexicographic ordering (in which
shorter programs are preferred to longer ones whenever their fitness is the
same or sufficiently similar). An approach which combines Pareto dominance and lexicographic ordering was proposed in (Panait and Luke, 2004).

9.3

Multiple Objectives via Dynamic and

Staged Fitness Functions

Often it is possible to rank multiple objectives based on some notion of

importance. In these cases, it is possible to use dynamic fitness functions

9.4 Multi-objective Optimisation via Operator Bias

81

which initially guide GP towards solutions that maximise the main objective. When enough of the population has reached reasonable levels in that
objective, the fitness function is modified so as to guide the population towards the optimisation of a second objective. In principle this process can
be iterated for multiple objectives. Of course, care needs to be taken to
ensure that the functionality reached with a set of previous fitness measures
is not wiped by the search for the optima of a later fitness function. This
can be avoided by making sure each new fitness function somehow includes
all the previous ones. For example, the fitness based on the new objectives
can be added to the pre-existing objectives with some appropriate scaling
factors.
A similar effect can be achieved via static, but staged, fitness functions.
These are staged in the sense that certain levels of fitness are only be made
available to an individual once it has reached a minimum acceptable performance on all objectives at the previous level. If each level represents one of
the objectives, individuals are then encouraged to evolve in directions that
ensure that good performance is achieved and retained on all objectives.
Koza et al. (1999) used this strategy when using GP for the evolution of
electronic circuits where many criteria, such as input-output performance,
power consumption, size, etc., must all be taken into account to produce
good circuits. Kalganova and Miller (1999) used Cartesian GP (see Section 7.2.3) to design combinational logic circuits. A circuits fitness was
given by a value between 0 and 100 representing the percentage of output
bits that were correct. If the circuit was 100% functional, then a further
component was added which represented the number of gates in the graph
that were not involved in the circuit. Since all individuals had the same
number of gates available in the Cartesian GP grid, this could be used to
minimise the number of gates actually used to solve the problem at hand.

9.4

MO GP via Operator Bias

While it is very common to use only modifications of the selection phase to

perform multi-objective search, it is also possible to combine MOO selection
with genetic operators exhibiting an inbuilt search bias which can steer the
algorithm towards optimising certain objectives.
In some sense the classical repair operators, which are used in constrained
optimisation to deal with hard constraints, are an extreme form of the idea
of using operators to help MOO search.1 More generally, it is possible to
imagine search operators with softer biases which favour the achievement
of one or more objectives. These can be the same or different from the
objectives that bias the selection of parents.
1 In combinatorial optimisation, repair operators are applied to invalid offspring to
modify them in such a way as to ensure a problems hard constraints are respected.

82

9 Multi-objective Genetic Programming

The pygmies and civil servants approach proposed in (Ryan, 1994, 1996)
combines the separation typical of Pareto-based approaches with biased
search operators. In this system two lists are built, one where individuals are ranked based on fitness and the other where individuals are ranked
based on a linear combination of fitness and size (i.e., a parsimonious fitness function). During crossover, the algorithm draws one parent from the
first list and the other from the second list. This can be seen as a form of
disassortative mating aimed at maintain diversity in the population. Another example of this kind is (Zhang and Rockett, 2005) where crossover
was modified so that an offspring is retained only if it dominates either of
its parents.
Furthermore, as discussed in Sections 5.2 and 11.3.2, there are several
mutation operators with a direct or indirect bias towards smaller programs.
This provides a pressure towards the evolution of more parsimonious solutions throughout a run.
As with the staged fitness functions discussed in the previous section,
it is also possible to activate operators with a known bias towards smaller
programs only when the main objective say a 100% correct solution has
been achieved. This was tested in (Pujol, 1999; Pujol and Poli, 1997), where
GP was used to evolve neural networks. After a 100% correct solution was
found, one hidden node of each network in the population was replaced by
a terminal, and the evolution process was resumed. This pruning procedure
was repeated until the specified number of generations had been reached.

Chapter 10

Fast and Distributed

Genetic Programming
Users of all artificial intelligence tools are always eager to extend the boundaries of their techniques, for example by attacking more and more difficult
problems. In fact, to solve hard problems it may be necessary to push GP
to the limit populations of millions of programs and/or long runs may be
necessary.
There are a number of techniques to speed up, parallelise and distribute
GP search. We start by looking at ways to reduce the number of fitness
evaluations or increase their effectiveness (Section 10.1) and ways to speed
up their execution (Section 10.2). We then look at the idea of running GP
in parallel (Section 10.3) and point out that faster evaluation is not the
only reason for doing so, as geographic distribution has advantages in its
own right. In Section 10.4 we describe masterslave parallel architectures
(Section 10.4.1), running GP on graphics hardware (Section 10.4.2) and
FPGAs (Section 10.4.3). Section 10.4.4 describes a fast method to exploit
the parallelism available on every computer. Finally, Section 10.5 concludes
this chapter with a brief discussion of distributed, even global, evolution of
programs.

10.1

Reducing Fitness Evaluations and/or

Increasing their Effectiveness

While admirers of linear GP will suggest that machine code GP is the ultimate in speed, all forms of GP can be made faster in a number of ways.
The first is to reduce the number of times a program is evaluated.
Many applications find the fitness of programs by running them on mul83

84

10 Fast and Distributed Genetic Programming

tiple training examples. The use of many examples provides an accurate

evaluation of a programs quality. However, ultimately the point of fitness
evaluation is to make a binary decision does this individual get a child or
not? The overwhelming proportion of GPs computational effort (or indeed
the effort in any evolutionary computation technique) goes into adjusting
the probability of this binary decision. However, it is not clear that a highprecision fitness evaluation is always necessary to decide well. Indeed, even
when the fitness evaluation is very accurate, most selection algorithms,1 being stochastic, inject noise into the decision of which points in the search
space to proceed from and which to abandon. In these cases, reducing the
number of times a program is evaluated is effectively an additional source
of noise. If a program has already demonstrated it works poorly compared
to the rest of the population on a fraction of the available training data, it
not likely to be chosen as a parent. Conversely, if it has already exceeded
many programs in the population after being tested on only a fraction of
the training set, then it is likely to be chosen as a parent (Langdon, 1998).
In either case, it is apparent that we do not gain much by running it on
the remaining training examples. Teller and Andre (1997) developed these
ideas into a useful algorithm called the rational allocation of trials.
As well as the computational cost, there are other negatives consequences
that come from using all the training data all the time, as doing so gives rise
to a static fitness function. In certain circumstances this may encourage the
population to evolve into a cul-de-sac where it is dominated by offspring of a
single initial program which did well on some fraction of the training cases,
but was unable to fit others. A static fitness function can create conditions
where good programs that perform moderately well on most portions of the
training data have lower fitness than those that do very well in only a few
small regions. With high selection pressure, it takes surprisingly little time
for the best individual to dominate the whole population.2
Gathercole and Ross (1994, 1997) investigated a number of ways of dynamically changing training samples,3 yielding a number of interacting effects. Firstly, by using only a subset of the available data, the GP fitness
evaluation took less time. Secondly, by changing which examples were being
1 Common selection algorithms include roulette wheel selection (Goldberg, 1989), SUS
(Baker, 1987) and tournament selection.
2 This is called the take over time (Goldberg, 1989). This can be formally analysed
(Blickle, 1996; Droste, Jansen, Rudolph, Schwefel, Tinnefeld, and Wegener, 2003), but
for tournament selection, a simple rule of thumb is often sufficient. If T is the tournament size, roughly logT (Pop size) generations are needed for the whole population to
become descendants of a single individual. If, for example, we use binary tournaments
(T = 2), then take over will require about ten generations for a population of 1,024.
Alternatively, if we have a population of one million (106 ) and use ten individuals in each
tournament (T = 10), then after about six generations more or less everyone will have
the same great6 great5 great4 great3 grand2 mother1 .
3 Siegel (1994) proposed a rather different implementation.

10.1 Reducing Fitness Evaluations/Increasing their Effectiveness

85

used over time, the evolving population saw more of the training data and so
was less liable to over fit a fraction of them. Thirdly, by randomly changing
the fitness function, it became more difficult for evolution to produce an
overspecialised individual which took over the population at the expense of
solutions which were viable on other parts of the training data. Dynamic
subset selection (DSS) appears to have been the most successful of Gathercoles suggested algorithms. It has been incorporated into Discipulus (see
page 63), and was recently used in a large data mining application (Curry,
Lichodzijewski, and Heywood, 2007).
Where each fitness evaluation may take a long time, it may be attractive to interrupt a long-running program in order to let others run. In GP
systems which allow recursion or contain iterative elements (Brave, 1996;
Langdon, 1998; Wilson and Heywood, 2007; Wong and Leung, 1996) it is
common to enforce a time limit, a limit on the number of instructions executed, or a bound on the number of times a loop is executed. Maxwell
(1994) proposed a solution to the question of what fitness to give to a program that has been interrupted. He allowed each program in the population
a quantum of CPU time. When the program used up its quantum it was
check-pointed.4 In Maxwells system, programs gained fitness as they ran,
i.e., each time a program correctly processed a fitness case, its fitness was
incremented. Tournament selection was then performed. If all members of
the tournament had used the same number of CPU quanta, then the fitter
program was the winner. If, however, one program had used less CPU than
the others (and had a lower fitness) then it was restarted and run until it
had used as much CPU as the others. Then fitnesses were compared in the
normal way.
Teller (1994) had a similar but slightly simpler approach: every individual in the population was run for the same amount of time. When the
allotted time elapsed a program was aborted and an answer extracted from
it, regardless of whether it had terminated or not. Teller called this an any
time approach. This suits graph systems like Tellers PADO (Section 7.2.2)
or linear GP (Chapter 7.1) where it is easy to designate a register as the
output register. The answer can then be extracted from this register or from
an indexed memory cell at any point (including whilst the programming is
running). Other any time approaches include (Spector and Alpern, 1995)
and (Langdon and Poli, 2008).
A simple technique to speed up the evaluation of complex fitness functions is to organise the fitness function into stages of progressively increasing
computational cost. Individuals are evaluated stage by stage. Each stage
contributes to the overall fitness of a program. However, individuals need
4 When a program is check-pointed, sufficient information (principally the program
counter and stack) is saved so that it can later be restarted from where it was stopped.
Many multi-tasking operating systems do something similar.

86

10 Fast and Distributed Genetic Programming

to reach a minimum fitness value in each stage in order for them to be

allowed to progress to the next stage and acquire further fitness. Often
different stages represent different requirements and constraints imposed on
solutions.
Recently, a sophisticated technique called backward chaining GP has
been proposed (Poli, 2005; Poli and Langdon, 2005a,b, 2006a). In GP and
other evolutionary algorithms which use tournament selection with small
tournament sizes, backward chaining can radically reduce the number of
fitness evaluations. Tournament selection randomly draws programs from
the population to construct tournaments, the winners of which are then selected. Although this process is repeated many times in each generation,
when the tournaments are small there is a significant probability that an
individual in the current generation is never chosen to become a member
of any tournament. By reordering the way operations are performed, backward chaining GP exploits this. It not only avoids fitness calculations for
individuals that are never included in a tournament, but can also achieve
higher fitness sooner.

10.2

Reducing Cost of Fitness with Caches

In computer hardware it is common to use data caches which automatically

hold copies of data locally in order to avoid the delays associated with fetching them from disk or over a network every time they are needed. This can
work well when a small amount of data is needed many times over a short
interval.
Caches can also be used to store results of calculations, thereby avoiding
the re-calculation of data (Handley, 1994). GP populations have enormous
amounts of common code (Langdon, 1998; Langdon and Banzhaf, 2005;
Langdon and Poli, 2008). This is, after all, how genetic search works: it
promotes the genetic material of fit individuals. So, typically in each generation we see many copies of successful code.
In many (but by no means all) GP systems, subtrees have no side-effects.
This means results pass through a programs root node in a well organised
and easy to understand fashion. Thus, if we remember a subtrees inputs
and output when it was run before, we can avoid re-executing code whenever
we are required to run the subtree again. Note that this is true irrespective
of whether we need to run the same subtree inside a different individual or
at a different time (i.e., a later generation). Thus, if we stored the output
with the root node, we would only need to run the subtree once for any
given set of inputs. Whenever the interpreter comes to evaluate the subtree,
it needs only to check if the subtrees root contains a cache of the values the
interpreter calculated last time, thus saving considerable computation time.
In order to achieve this, however, we need to overcome a problem: not

10.2 Reducing Cost of Fitness with Caches

87

only must the answer be stored, but the interpreter needs to know that the
subtrees inputs are the same too. The common practices of GP come to our
aid here. Usually every tree in the population is run on exactly the same
inputs for each of the fitness cases. Thus, for a cache to work, the interpreter
does not need to know a trees inputs in detail, it need only know which of
the fixed set of test cases was used.
A simple means of implementing this type of cache is to store a vector of
values returned by each subtree for each of the test cases. Whenever a subtree is created (i.e., in the initial generation, by crossover or by mutations)
the interpreter is run and the cache of values for its root node is set. Note
this is recursive, so caches can also be calculated for subtrees within it at
the same time. Now, when the interpreter is run and comes to a subtrees
root node, it will simply retrieve the value it calculated earlier, using the
test cases number as an index into the cache vector.
If a subtree is created by mutation, then its cache of values will be
initially empty and will have to be calculated. However, this costs no more
than it would without caches.
When code is inserted into an existing tree, be it by mutation or
crossover, the chance that the new code behaves identically to the old code
is normally very small. This means that the caches of every node between
the new code and the root node may be invalid. The simplest solution is
to re-evaluate them all. This may sound expensive, but the caches in all
the other parts of the individual remain valid and can be used when the
cache above them is re-evaluated. Thus, in effect, if the crossed over code is
inserted at depth d, only d nodes need to be evaluated.
The whole question of monitoring how effective individual caches are,
what their hit-rates are, etc. has been little explored. In practice, impressive
savings have been achieved by simple implementations, with little monitoring and rudimentary garbage collection. Recent analysis (Ciesielski and Li,
2004; Dignum and Poli, 2007; Langdon and Poli, 2002; Poli et al., 2007)
has shown that GP trees tend not to have symmetric shapes, and many
leaves are very close to the root. This provides a theoretical explanation for
why considerable computational saving can be made by using fitness caches.
While it is possible to use hashing schemes to efficiently find common code,
in practice assuming that common code only arises because it was inherited
from the same location (e.g., by crossing over) is sufficient.
As well as the original Directed acyclic graph (DAG) implementation
(Handley, 1994) other work includes (Ciesielski and Li, 2004; Keijzer, 1996;
McPhee, Hopper, and Reierson, 1998; Yangiya, 1995). While so far we have
only considered programs where no side effects take place, there are cases
where caching can be extended outside this domain. For example, Langdon
(1998) used fitness caches in evolved trees with side effects by exploiting
syntax rules about where in the code the side-effects could lie.

88

10.3

10 Fast and Distributed Genetic Programming

Parallel and Distributed GP are Not

Equivalent

There are two important aspects of parallel evolutionary algorithms which

are equally important but are often confused. The first is the traditional
aspect of parallel computing. That is, we port an existing algorithm onto
a supercomputer so that it runs faster. The second aspect comes from the
biological inspiration for evolutionary computation.
In nature everything happens in parallel. Individuals succeed or fail in
producing and raising children at the same time as other members of their
species. These individuals are spread across oceans, lakes, rivers, plains,
forests, mountain chains, etc. It was this geographic spread that led Wright
(1932) to propose that geography and changes to it are of great importance
to the formation of new species and, so, to natural evolution as a whole.
Suppose a species occupies a range of hills. Individuals need not be able
to move from one end of the range to another in their lifetime, but their
descendents might. Wright (1932) proposed a mathematical model that can
predict the amount of mixing between descendents across the entire range
is needed to keep the whole population together as a single species. Based
on his model, he predicted that only a few migrants per generation between
hill tops are sufficient.
Now suppose the sea level rises. What was once a continuous range
of hills becomes a chain of islands. Suppose members of this species have
limited ability to swim. If the islands are close and the ocean currents are
sometimes favourable, it may be that every year a few individuals cross
between neighbouring islands. This may be enough to constrain diversification and allow the population to remain a single species. However, if the
gaps between island become larger, the chance of an individual occasionally
crossing the sea and breeding becomes remote. On each island, then, the
sub-populations begin to diverge and over time new species, specific to each
island, are formed (Darwin, 1859).
In nature, changes in conditions across regions can lead to corresponding differences in spatially distributed populations. Sometimes this can lead
to new species, as in the example above. In other cases the variation can
be gradual enough that there is no clear delineation that could be called a
species boundary, but geographically distant individuals are unable or unwilling to mate, fulfilling a key property of different species. A particularly
dramatic example of this is a ring species. The Larus gulls, for example, live
along a ring that roughly follows the Arctic Circle. With one exception the
variants can interbreed all along its range, despite often having differences
significant enough that they have received different names. The key exception is in Europe, where the ends of the range meet. There the Herring
Gull (Larus argentatus) and the Lesser Black-backed Gull (Larus fuscus)

10.4 Running GP on Parallel Hardware

89

intermingle, but rarely interbreed.

The topology of the landscape is often a strong determiner of the spatial
structure of a population. A large, fairly homogenous region, for example,
might give rise to a species being spatially differentiated in two dimensions
due to distance and climate. A river, on the other hand, may give rise to
a linear distribution, especially if there are structures like water falls that
restrict migration, and a river basin with several tributaries could lead to a
tree structure.
In evolutionary computation we can choose whether we want to model
some form of geography. We can run GP on parallel hardware so as to speed
up runs, but without introducing a notion of proximity that limits which
individuals are allowed to mate. Alternatively, we can model some form of
geography, introducing spatial structure as a result.
In the following two sections we will discuss both ideas. It is important
to note, however, that one does not need to use parallel hardware to use geographically distributed GP populations. Although parallel hardware naturally lends itself to the implementation of physically-distributed populations,
one can obtain similar benefits by using logically-distributed populations in
a single machine.

10.4

Running GP on Parallel Hardware

In contrast to much of computer science, evolutionary computation can be

readily run on parallel computer hardware; indeed it is embarrassingly parallel (Andre and Koza, 1998). For example, when Openshaw and Turton
(1994) ran GP on a Cray supercomputer they obtained about 30% of its
theoretical peak performance, embarrassing their supercomputer savvy colleagues who rarely got better than a few percent out of it.
In Sections 10.4.110.4.3 we look at three ways of running GP on parallel hardware. Section 10.4.4 shows how to get 32 parallel operations from
standard hardware.

10.4.1

Masterslave GP

If the objective is purely to speed up runs, we may want our parallel GP to

work exactly the same as it did on a single computer. This is possible, but
to achieve it we have to be very careful to ensure that, even if some parts of
the population are evaluated more quickly, parallelisation does not change
how we apply selection and which GP individual crosses over with which.
Probably the easiest way to implement this is the masterslave model.
In the masterslave model (Oussaid`ene, Chopard, Pictet, and Tomassini,
1997) breeding, selection crossover, mutation etc. occur just as they would
on a single computer and only fitness evaluation is spread across a network

90

10 Fast and Distributed Genetic Programming

of computers. Each GP individual and its fitness cases are sent across the
network to a different compute node. The central node waits for the compute
nodes to return their individuals fitnesses. Since individuals and fitness
values are typically stored in small data structures, this can be quite efficient
since transmission overheads are limited.
The central node is an obvious bottleneck. Also, a slow compute node
or a lengthy fitness case will slow down the whole GP population, since
eventually its result will be needed before moving onto the next generation.

10.4.2

GP Running on GPUs

Modern PC graphics cards contain powerful graphics processing units

(GPUs) including a large number of computing components. For example, it is not atypical to have 128 streaming processors on a single PCs
graphics card. In the last few years there has been an explosion of interest
in porting scientific or general purpose computation to mass market graphics
cards (Owens, Luebke, Govindaraju, Harris, Kruger, Lefohn, and Purcell,
2007).
Indeed, the principal manufactures (nVidia and ATI) claim faster than
Moores Law (Moore, 1965) increase in performance, suggesting that GPU
floating point performance will continue to double every twelve months,
rather than the 1824 months observed for electronic circuits in general and
personal computer CPUs in particular. In fact, the apparent failure of PC
CPUs to keep up with Moores law in the last few years makes GPU computing even more attractive. Even todays bottom-of-the-range GPUs greatly
exceed the floating point performance of their hosts CPU. However, this
speed comes at a price, since GPUs provide a restricted type of parallel processing, often referred to a single instruction multiple data (SIMD) or single
program multiple data (SPMD). Each of the many processors simultaneously
runs the same program on different data items.
There have been a few genetic programming experiments with GPUs
(Chitty, 2007; Ebner, Reinhardt, and Albert, 2005; Harding and Banzhaf,
2007; Langdon and Banzhaf, 2008; Langdon and Harrison, 2008; Loviscach
and Meyer-Spradow, 2003; Meyer-Spradow and Loviscach, 2003; Reggia,
Tagamets, Contreras-Vidal, Jacobs, Weems, Naqvi, Winder, Chabuk, Jung,
and Yang, 2006). So far, in GP, GPUs have just been used for fitness evaluation.
Harding and Banzhaf (2007) used the Microsoft research GPU development DirectXTM tools to compile (using a technique originally developed by
Harris and Buxton (1996)) a whole population of Cartesian GP network programs into a single GPU program which was loaded onto a laptops GPU to
run the fitness cases. Chitty (2007) used a conversion technique, somewhat
like an interpreter, to automatically convert each GP tree into a program

PCI Express

TA TA TA TA

TF TF TF TF

TF TF TF TF

TA TA TA TA

TF TF TF TF

TF TF TF TF
L1

L1

L2 CACHE

64 MB 64 MB

memory
32bits 32bits

ROP

L1

TF TF TF TF

TF TF TF TF

TA TA TA TA

SP SP SP SP

SP SP SP SP

SP SP SP SP

SP SP SP SP

L1

64 MB 64 MB

L1

L2 CACHE

L1

TF TF TF TF

TF TF TF TF

TA TA TA TA

SP SP SP SP

SP SP SP SP

SP SP SP SP

SP SP SP SP

64 MB 64 MB

memory
32bits 32bits

ROP

L1

TF TF TF TF

TF TF TF TF

TA TA TA TA

SP SP SP SP

SP SP SP SP

SP SP SP SP

SP SP SP SP

L2 CACHE

L1

memory
32bits 32bits

ROP

L1

TF TF TF TF

TF TF TF TF

TA TA TA TA

SP SP SP SP

SP SP SP SP

SP SP SP SP

SP SP SP SP

Thread Execution Manager

L2 CACHE

L1

64 MB 64 MB

memory
32bits 32bits

ROP

L1

TF TF TF TF

TF TF TF TF

TA TA TA TA

SP SP SP SP

SP SP SP SP

SP SP SP SP

SP SP SP SP

Thread Processor

Figure 10.1: nVidia 8800 Block diagram. The 128 1360 MHz Stream Processors are arranged in 16 blocks of 8. Blocks
share 16 KB memory (not shown), an 8/1 KB L1 cache, 4 Texture Address units and 8 Texture Filters. The 664 bit
bus (dashed) links off chip RAM at 900 MHz. (Since there are two chips for each of the six off-chip memory banks, the
bus is effectively running at up to 1800 Mhz per bank.) There are 6 Raster Operation Partitions. (nVidia, 2007).

64 MB 64 MB

L2 CACHE

64 MB 64 MB

ROP

memory
32bits 32bits

L2 CACHE

L1

TF TF TF TF

TF TF TF TF

TA TA TA TA

SP SP SP SP

SP SP SP SP

SP SP SP SP

SP SP SP SP

memory
32bits 32bits

ROP

L1

SP SP SP SP

SP SP SP SP

L1

SP SP SP SP

SP SP SP SP

L1

SP SP SP SP

SP SP SP SP

L1

SP SP SP SP

SP SP SP SP

Input Assembler

PC

10.4 Running GP on Parallel Hardware

91

92

10 Fast and Distributed Genetic Programming

that could be compiled for the GPU on the host PC. The compiled programs were transferred one at a time to a GPU for fitness evaluation. Both
groups obtained impressive speedups by running many test cases in parallel.
Langdon and Banzhaf (2008) and Langdon and Harrison (2008) created
a SIMD interpreter (Juille and Pollack, 1996) using RapidMinds GNU C++
OpenGL framework to simultaneously run up to a quarter of a million GP
trees on an NVIDIA GPU (see Figure 10.1).5 As discussed in Section 7.1.2,
GP trees can be linearised. This avoids pointers and yields a very compact
data structure; reducing the amount of memory needed in turn facilitates
the use of large populations. To avoid recursive calls in the interpreter,
Langdon used reverse polish notation (RPN), i.e., a post-fix rather than
a pre-fix notation. Only small modifications are needed to crossover and
mutation so that they act directly on the RPN expressions. This means the
same representation is used on both the host and the GPU. Almost a billion
GP primitives can be interpreted by a single graphics card per second. In
both Cartesian and tree-based GP the genetic operations are done by the
host CPU. Wong, Wong, and Fok (2005) showed, for a genetic algorithm,
these too can be done by the GPU.
Although each of the GPUs processors may be individually quite fast
and the manufacturers claim huge aggregate FLOPS ratings, the GPUs are
optimised for graphics work. In practice, it is hard to keep all the processors
fully loaded. Nevertheless 30 GFLOPS has been achieved (Langdon and
Harrison, 2008). Given the differences in CPU and GPU architectures and
clock speeds, often the speedup from using a GPU rather than the host
CPU is the most useful statistic. This is obviously determined by many
factors, including the relative importance of amount of computation and
size of data. The measured RPN tree speedups were 7.6-fold (Langdon and
Harrison, 2008) and 12.6-fold (Langdon and Banzhaf, 2008).

10.4.3

GP on FPGAs

Field programmable gate arrays (FPGAs) are chips which contain large arrays of simple logic processing units whose functionality and connectivity
can be changed via software in microseconds by simply writing a configuration into a static memory. Once an FPGA is configured it can update
all of its thousands of logic elements in parallel at the clock speed of the
circuit. Although an FPGAs clock speed is often an order of magnitude
slower than that of a modern CPU, its massive parallelism makes it a very
powerful computational device. Because of this and of their flexibility there
has been significant interest in using FPGAs in GP.
Work has ranged from the use of FPGAs to speed up fitness evaluation
5 Bigger populations, e.g. five million programs (Langdon and Harrison, 2008), are
possible by loading them onto the GPU in 256k units.

10.5 Geographically Distributed GP

93

(Koza, Bennett, Hutchings, Bade, Keane, and Andre, 1997; Seok, Lee, and
Zhang, 2000) to the definition of specialised operators (Martin and Poli,
2002). It is even possible to implement a complete GP on FPGAs, as suggested in (Heywood and Zincir-Heywood, 2000; Martin, 2001, 2002; Sidhu,
Mei, and Prasanna, 1998). A massively parallel GP implementation has also
been proposed by Eklund (2001, 2004) although to date all tests with that
architecture have only been performed in simulation.

10.4.4

Sub-machine-code GP

We are nowadays so used to writing programs using high level sequential

languages that it is very easy to forget that, underneath, computers have a
high degree of parallelism. Internally, CPUs are made up of bit-slices which
make it possible for the CPU to process all of the bits of the operands of an
instruction in one go, in a single clock tick.
Sub-machine-code GP (SMCGP) (Poli and Langdon, 1999) is a technique
to speed up GP and to extend its scope by exploiting the internal parallelism
of sequential CPUs. In Boolean classification problems, SMCGP allows the
parallel evaluation of 32 or 64 (depending on the CPUs word size) fitness
cases per program execution, thereby providing a significant speed-up. This
has made it possible to solve parity problems with up to 4 million fitness
cases (Poli and Page, 2000). SMCGP has also been applied with success
in binary image classification problems (Adorni, Cagnoni, and Mordonini,
2002; Quintana, Poli, and Claridge, 2003). The technique has also been
extended to process multiple fitness cases per program execution in continuous symbolic regression problems where inputs and outputs are real-valued
numbers (Poli, 1999b).

10.5

Geographically Distributed GP

Unless some type of synchronisation is imposed, the parallel forms of GP

in which different parts of a population are evolved by different processing
elements will not be running the same algorithm as the standard singleCPU version of GP. Therefore, almost certainly, different parallelisations
will produce different answers. However, as we discussed in Section 10.3,
this is not necessarily a bad thing.
Parallelisation itself can bring benefits similar to those hypothesised in
natural populations by Wright (1932). In particular, the population is often divided into semi-independent sub-populations called demes (Collins,
1992; Dhaeseleer and Bluming, 1994; Langdon, 1998; Popovici and De Jong,
2006). The flow of genetic material between demes is restricted by limiting
the exchange of individuals between them. The limit can be on the number
of individuals that are allowed to migrate per generation. Alternatively the

94

10 Fast and Distributed Genetic Programming

(a)

(b)

Figure 10.2: Spatially structured GP populations. (a) Toroidal grid of

demes, where each deme (a node) contains a sub-population, and demes
periodically exchange a small group of high-fitness individuals using a grid
of communication channels. (b) Fine-grained distributed GP, where each
grid cell contains one individual and where the selection of a mating partner
for the individual in the centre cell is performed by executing a tournament
among randomly selected individuals (e.g., the individuals shaded) in its
3 3 neighbourhood.

demes may be considered to be arranged in a geographical topology that

constrains which demes can trade individuals. For example, it may be that
with limited migration between compute nodes, the evolved populations on
adjacent nodes will diverge, and that this increased diversity may lead to
better solutions. Fernandez, Tomassini, and Vanneschi (2003), for example, report that distributing individuals between subpopulations offers an
advantage in terms of quality of solutions and computational effort.
When Koza first started using GP on a network of Transputers (Andre
and Koza, 1996), Andre experimentally determined the best migration rate
for their problem. He suggested Transputers arranged in an asynchronous
2-D toroidal square grid (such as the one in Figure 10.2a) should exchange
2% of their population with their four neighbours.
Densely connected grids have been widely adopted in parallel GP. Usually they allow innovative partial solutions to spread quickly. However, the
GA community reported better results from less connected topologies, such
as arranging the compute nodes populations in a ring, so that they could
transfer genes only between themselves and their two neighbours (Stender,
1993). Potter (1997) argues in favour of spatial separation in populations
and fine-grained distributed forms of GP (see Figure 10.2b). Whitley (2001)
gives some guidance on parallel genetic algorithms.

10.5 Geographically Distributed GP

95

Figure 10.3: A globally distributed GP system (Langdon, 2005a). The

server is the centre of the star architecture, with the lines connecting it
to users around the world. The users evolved snowflake patterns using a
continuously evolving L-System, and their (subjective) preferences provided
the fitness measure used to drive the system.
While many have looked enviously at Kozas 1000 node Beowulf cluster
(Sterling, 1998) and other supercomputer realisations of GP (Bennett, Koza,
Shipman, and Stiffelman, 1999; Juille and Pollack, 1996), a supercomputer is
often not necessary. Many businesses and research centres leave computers
permanently switched on. During the night their computational resources
tend to be wasted. This computing power can easily and efficiently be
used to execute distributed GP runs overnight. Typically, GP does not
demand a high performance bus to interconnect the compute nodes, and
so existing office Ethernet networks are often sufficient. While parallel GP
systems can be implemented using MPI (Walker, 2001) or PVM (Fernandez,
Sanchez, Tomassini, and Gomez, 1999), the use of such tools is not necessary:
simple Unix commands and port-to-port HTTP is sufficient (Poli, Page,
and Langdon, 1999). The population can be split and stored on modest
computers. With only infrequent interchange of parts of the population
or fitness values little bandwidth is needed. Indeed a global population
spread via the Internet (Chong and Langdon, 1999; Draves, 2006; Klein and
Spector, 2007; Langdon, 2005a), `
a la seti@home, is perfectly feasible (see
Figure 10.3).
Other parallel GPs include (Cheang, Leung, and Lee, 2006; Folino, Pizzuti, and Spezzano, 2003; Gustafson and Burke, 2006; Klein and Spector,
2007; Tanev, Uozumi, and Akhmetov, 2004).

Chapter 11

GP Theory and its

Applications
Most of this book is about the mechanics of GP and its practical use for
solving problems. In fact, as will become clear in Chapter 12, GP has
been remarkably successful as a problem-solving and engineering tool. One
might wonder how this is possible, given that GP is a non-deterministic
algorithm, and as a result its behaviour varies from run to run. It is also a
complex adaptive system which sometimes shows intricate and unexpected
behaviours (such as bloat). Thus it is only natural to be interested in GP
from the scientific point of view. That is, we want to understand why can
GP solve problems, how it does it, what goes wrong when it cannot, what are
the reasons for certain undesirable behaviours, what can we do to get rid of
them without introducing new (and perhaps even less desirable) problems,
and so on.
GP is a search technique that explores the space of computer programs.
The search for solutions to a problem starts from a group of points (random
programs) in this search space. Those points that are above average quality
are then used to generate a new generation of points through crossover,
mutation, reproduction and possibly other genetic operations. This process
is repeated over and over again until a stopping criterion is satisfied. If we
could visualise this search, we would often find that initially the population
looks like a cloud of randomly scattered points, but that, generation after
generation, this cloud changes shape and moves in the search space. Because
GP is a stochastic search technique, in different runs we would observe
different trajectories. If we could see regularities, these might provide us
with a deep understanding of how the algorithm is searching the program
space for the solutions, and perhaps help us see why GP is successful in
finding solutions in certain runs and unsuccessful in others. Unfortunately,
97

98

11 GP Theory and its Applications

it is normally impossible to exactly visualise the program search space due

to its high dimensionality and complexity, making it that much harder to
understand.
An alternative approach to better understanding the dynamics of GP is
to study mathematical models of evolutionary search. There are a number
of cases where this approach has been very successful in illuminating some
of the fundamental processes and biases in GP systems. In this chapter we
will review several theoretical approaches to understanding GP, including
mathematical models of GP (Section 11.1), analyses of the structure of GP
search spaces (Section 11.2), and the use of theory to understand and combat
the chronic problem of bloat in a principled fashion (Section 11.3).

11.1

Mathematical Models

Schema theories are among the oldest and the best known models of evolutionary algorithms (Holland, 1992; Whitley, 1994). Schema theories are
based on the idea of partitioning the search space into subsets, called
schemata. They are concerned with modelling and explaining the dynamics
of the distribution of the population over the schemata. Modern genetic
algorithm schema theory (Stephens and Waelbroeck, 1997, 1999) provides
exact information about the distribution of the population at the next generation in terms of quantities measured at the current generation, without
having to actually run the algorithm.
The theory of schemata in GP has had a difficult childhood. Some excellent early efforts led to different worst-case-scenario schema theorems (Altenberg, 1994; Koza, 1992; OReilly and Oppacher, 1994b; Poli and Langdon,
1997; Rosca, 1997; Whigham, 1995). Only very recently have the first exact schema theories become available (Poli, 2000a,b, 2001a) which give exact
formulations (rather than lower bounds) for the expected number of individuals sampling a schema at the next generation. Initially (Poli, 2000b, 2001a),
these exact theories were only applicable to GP with one-point crossover (see
Section 5.3). However, more recently they have been extended to the class of
homologous crossovers (Poli, McPhee, and Rowe, 2004) and to virtually all
types of crossovers that swap subtrees (Poli and McPhee, 2003a,b), including
standard GP crossover with and without uniform selection of the crossover
points (Section 2.4), one-point crossover, context-preserving crossover and
size-fair crossover (which have been described in Section 5.3), as well as
more constrained forms of crossover such as strongly-typed GP crossover
(see Section 6.2.2), and many others.
Other models of evolutionary algorithms include models based on
Markov chain theory (e.g. (Davis and Principe, 1993; Nix and Vose, 1992))
and on statistical mechanics (e.g. (Pr
ugel-Bennett and Shapiro, 1994)).
Markov models have been applied to GP (Mitavskiy and Rowe, 2006; Poli
et al., 2004; Poli, Rowe, and McPhee, 2001), but so far they have not been

11.2 Search Spaces

99

developed as fully as the schema theory model.

Exact mathematical models of GP are probabilistic descriptions of the
operations of selection, reproduction, crossover and mutation. They explicitly represent how these operations determine which areas of the program
space will be sampled by GP, and with what probability. These models treat
the fitness function as a black box, however. That is, there is no representation of the fact that in GP, unlike in other evolutionary techniques, the
fitness function involves the execution of computer programs on a variety of
inputs. In other words, schema theories and Markov chains do not tell us
how fitness is distributed in the search space. Yet, without this information,
we have no way of closing the loop and fully characterising the behaviour of a
GP systems which is always the result of the interaction between the fitness
function and the search biases of the representation and genetic operations
used in the system.

11.2

Search Spaces

The characterisation of the space of computer programs explored by GP has

been another main topic of theoretical research (Langdon and Poli, 2002).
Of course results describing the space of all possible programs are widely
applicable, not only to GP and other search-based automatic programming
techniques, but also to many other areas ranging from software engineering
to theoretical computer science.
In this category are theoretical results showing that the distribution of
functionality of non Turing-complete programs approaches a limit as program length increases. That is, although the number of programs of a
particular length grows exponentially with length, beyond a certain threshold the fraction of programs implementing any particular functionality is
effectively constant. For example, in Figure 11.1 we plot the proportion of
binary program trees composed of NAND gates which implement each of the
3
22 = 256 Boolean functions of three inputs. Notice how, as the length of
programs increases, the proportion of programs implementing each function
approaches a limit.
This does not happen by accident. There is a very substantial body of
empirical evidence indicating that this happens in a variety of other systems.
In fact, there are also mathematical proofs of these convergence results for
two important forms of programs: Lisp (tree-like) S-expressions (without
side effects) and machine code programs without loops (Langdon, 2002a,b,
2003a,b, 2005b; Langdon and Poli, 2002). That the limiting distribution of
functionality reaches a limit as program length increases was also proven for
a variety of other non-Turing complete computers and languages, including:
a) cyclic (increment, decrement and NOP), b) bit flip computer (flip bit
and NOP), c) any non-reversible computer, d) any reversible computer,

100

11 GP Theory and its Applications

0.1
0.01
0.001
0.0001
1e-05
1e-06

255

1e-07

201
0

10

20

30

40

50

Three-Input Boolean equivalence class

60

70

80

31

63

91

151
127

Size

Figure 11.1: Proportion of NAND trees that yield each three-input functions. As circuit size increases the distribution approaches a limit.
e) CCNOT (Toffoli gate) computer, f) quantum computers, g) the average
computer and h) AND, NAND, OR, NOR expressions.
Recently, (Langdon and Poli, 2006; Poli and Langdon, 2006b) started extending these results to Turing complete machine code programs. For this
purpose, a simple, but realistic, Turing complete machine code language,
T7, was considered. It includes: directly accessed bit addressable memory,
an addition operator, an unconditional jump, a conditional branch and four
copy instructions. A mathematical analysis of the halting process based on
a Markov chain model of program execution and halting was performed.
The model can be used to estimate, for any given program length, important quantities, such as the halting probability and the run time of halting
programs. This showed a scaling law indicating
that the halting probabil
number
ity for programs of length L is of order 1/ L, while the expected

of instructions executed by halting programs is of order L. In contrast

to many proposed Markov models, this can be done very efficiently, making it possible to compute these quantities for programs of tens of million
instructions in a few minutes. Experimental results confirmed the theory.

11.3 Bloat

11.3

101

Bloat

Starting in the early 1990s, researchers began to notice that in addition to

progressively increasing their mean and best fitness, GP populations also
showed certain other dynamics. In particular, it was noted that very often
the average size (number of nodes) of the programs in a population, after a
certain number of generations in which it was largely static, at some point
would start growing at a rapid pace. Typically the increase in program size
was not accompanied by any corresponding increase in fitness. The origin
of this phenomenon, which is known as bloat, has effectively been a mystery
for over a decade.
Note that there are situations where one would expect to see program
growth as part of the process of solving a problem. For example, GP runs
typically start from populations of small random programs, and it may be
necessary for the programs to grow in complexity for them to be able to
comply with all the fitness cases (a situation which often arises in continuous
symbolic regression problems). So, we should not equate growth with bloat
and we should define bloat as program growth without (significant) return in
terms of fitness.
Bloat is not only surprising, it also has significant practical effects: large
programs are computationally expensive to evolve and later use, can be hard
to interpret, and may exhibit poor generalisation. For these reasons bloat
has been a subject of intense study in GP. Over the years, many theories
have been proposed to explain various aspects of bloat, and while great
strides have been made, we still lack a single, universally-accepted unifying
theory to explain the broad range of empirical observations. We review the
key theoretical results on bloat in Section 11.3.1.
While discussions on the causes of bloat were going on, practitioners have
still had to face the reality of combating bloat in their runs. Consequently,
a variety of effective practical techniques have been proposed to counteract
bloat. We review these in Section 11.3.2, where we will particularly focus on
the parsimony pressure method (Koza, 1992; Zhang and M
uhlenbein, 1993,
1995; Zhang et al., 1997), which is perhaps the simplest and most frequently
used method to control bloat in genetic programming.

11.3.1

Bloat in Theory

As mentioned above, there are several theories of bloat. Let us start by

looking at three of the oldest ones: the replication accuracy theory, the
removal bias theory and the nature of program search spaces theory.

102

11 GP Theory and its Applications

Three Classic Explanations for Bloat

The replication accuracy theory (McPhee and Miller, 1995) states that the
success of a GP individual depends on its ability to have offspring that are
functionally similar to the parent. As a consequence, GP evolves towards
(bloated) representations that increase replication accuracy.
The nodes in a GP tree can often be crudely categorised into two classes:
active code and inactive code. Roughly speaking, inactive code is code
that is not executed, or is executed but its output is then discarded. All
remaining code is active code. The removal bias theory (Soule and Foster,
1998a) observes that inactive code in a GP tree tends to be low in the tree,
residing, therefore, in smaller-than-average-size subtrees. Crossover events
excising inactive subtrees produce offspring with the same fitness as their
parents. On average the inserted subtree is bigger than the excised one, so
such offspring are bigger than average while retaining the fitness of their
parent, leading ultimately to growth in the average program size.
Finally, the nature of program search spaces theory (Langdon and Poli,
1997; Langdon, Soule, Poli, and Foster, 1999) predicts that above a certain
size, the distribution of fitnesses does not vary with size. Since there are more
long programs, the number of long programs of a given fitness is greater than
the number of short programs of the same fitness. Over time GP samples
longer and longer programs simply because there are more of them.

Executable Models of Bloat

The explanations for bloat provided by these three theories are largely qualitative. There have, however, been some efforts to mathematically formalise
and verify these theories. For example, Banzhaf and Langdon (2002) defined
an executable model of bloat where only the fitness, the size of active code and
the size of inactive code were represented (i.e., there was no representation of
program structures). Fitnesses of individuals were drawn from a bell-shaped
distribution, while active and inactive code lengths were modified by a sizeunbiased mutation operator. Various interesting effects were reported which
are very similar to corresponding effects found in GP runs. Rosca (2003) proposed a similar, but slightly more sophisticated model which also included
an analogue of crossover. This provided further interesting evidence.
A strength of these executable models is their simplicity. A weakness is
that they suppress or remove many details of the representation and operators typically used in GP. This makes it difficult to verify if all the phenomena observed in the model have analogues in GP runs, and if all important
behaviours of GP in relation to bloat are captured by the model.

11.3 Bloat

103

Size Evolution Equation

In (Poli, 2001b; Poli and McPhee, 2003b), a size evolution equation for genetic programming was developed, which provided an exact formalisation of
the dynamics of average program size. The original equation was derived
from the exact schema theory for GP, and expressed mean program size as
a function of the size and selection probabilities of particular schemata representing program shapes. The equation has recently been simplified (Poli
and McPhee, 2008b) giving:
X
E[(t + 1)] =
` p(`, t),
(11.1)
`

where (t + 1) is the mean size of the programs in the population at generation t + 1, E is the expectation operator, ` is a program size, and p(`, t)
is the probability of selecting programs of size ` from the population in
generation t.
This equation can be rewritten in terms of the expected change in average
program size as:
X
E[(t + 1) (t)] =
` (p(`, t) (`, t)),
(11.2)
`

where (`, t) is the proportion of programs of size ` in the current generation. Both equations apply to a GP system with selection and any form of
symmetric subtree crossover.1
Note that Equations (11.1) and (11.2) do not directly explain bloat. They
are, however, important because they constrain what can and cannot happen size-wise in GP populations. Any explanation for bloat (including the
theories summarised above) has to agree with Equations (11.1) and (11.2).
In particular, Equation (11.1) predicts that, for symmetric subtreeswapping crossover operators, the mean program size evolves as if selection
only was acting on the population. This means that if there is a change in
mean size (bloat, for example) it must be the result of some form of positive
or negative selective pressure on some or all of the length classes `. Equation (11.2) shows that there can be bloat only if the selection probability
p(`, t) is different from the proportion (`, t) for at least some `. In particular, for bloat to happen there will have to be some small `s for which
p(`, t) < (`, t) and also some bigger `s for which p(`, t) > (`, t) (at least
on average).

1 In a symmetric operator the probability of selecting particular crossover points in the

parents does not depend on the order in which the parents are drawn from the population.

104

11 GP Theory and its Applications

Crossover Bias Theory of Bloat

We conclude this review on theories of bloat with a recent explanation for
bloat called the crossover bias theory (Dignum and Poli, 2007; Poli et al.,
2007). This is based on and is consistent with the size evolution equation
(Equation 11.1).
On average, each application of subtree crossover removes as much genetic material as it inserts; consequently crossover on its own does not produce growth or shrinkage. While the mean program size is unaffected, however, higher moments of the distribution are. In particular, crossover pushes
the population towards a particular distribution of program sizes, known
as a Lagrange distribution of the second kind, where small programs have a
much higher frequency than longer ones. For example, crossover generates a
very high proportion of single-node individuals. In virtually all problems of
practical interest, however, very small programs have no chance of solving
the problem. As a result, programs of above average size have a selective
advantage over programs of below average size, and the mean program size
increases.
Because crossover will continue to create small programs, which will then
be ignored by selection (in favour of the larger programs), the increase in
average size will continue generation by generation.

11.3.2

Bloat Control in Practice

Numerous empirical techniques have been proposed to control bloat (Langdon et al., 1999; Soule and Foster, 1998b). We cannot look at them all.
However, we briefly review some of the most important.
Size and Depth Limits
Rather naturally, the first and simplest method to control code growth is the
use of hard limits on the size or depth of the offspring programs generated
by the genetic operators.
Many implementations of this idea (e.g., (Koza, 1992)) apply a genetic
operator and then check whether the offspring is beyond the size or depth
limit. If it isnt, the offspring enters the population. If, instead, the offspring exceeds the limit, one of the parents is returned. Obviously, this
implementation does not allow programs to grow too large. However, there
is a serious problem with this way of applying size limits, or more generally,
constraints to programs: parent programs that are more likely to violate a
constraint will tend to be copied (unaltered) more often than programs that
dont. That is, the population will tend to be filled up with programs that
nearly infringe the constraint, which is typically not what is desired.

11.3 Bloat

105

It is well known, for example, that depth thresholds lead to the population filling up with very bushy programs where most branches reach the
depth limit (being effectively full trees). On the contrary, size limits produce
populations of stringy programs which tend to all approach the size limit.
See (Crane and McPhee, 2005; McPhee, Jarvis, and Crane, 2004) for more
on the impact of size and depth limits, and the differences between them.
The problem can be fixed by not returning parents if the offspring violates
a constraint. This can be realised with two different strategies. Firstly, one
can just return the oversize offspring, but give it a fitness of 0, so that
selection will get rid of it at the next generation. Secondly, one can simply
declare the genetic operation failed, and try again. This can be done in two
alternative ways: a) the same parent or parents are used again, but new
mutation or crossover points are randomly chosen (which can be done up
to a certain number of times before giving up on those parents), or b) new
parents are selected and the genetic operation is attempted again.
If a limit is used, programs must not be so tightly constrained that they
cannot express any solution to the problem. As a rule of thumb, one should
try to estimate the size of the minimum possible solution (using the terminals
and functions given to GP) and add some percentage (e.g., 50-200%) as a
safety margin. In general, however, it may be hard to heuristically come up
with good limits, so some trial and error may be required. Alternatively,
one can use one of the many techniques that have been proposed to adjust
size limits during runs. These can be both at the level of individuals and the
population. See for example the work by Silva and Almeida (2003); Silva
and Costa (2004, 2005a,b); Silva, Silva, and Costa (2005).
Anti-bloat Genetic Operators
One can control bloat by using genetic operators which directly or indirectly
have an anti-bloat effect.
Among the most recent bloat-control methods are size fair crossover
and size fair mutation (Crawford-Marks and Spector, 2002; Langdon, 2000).
These work by constraining the choices made during the execution of a
genetic operation so as to actively prevent growth. In size-fair crossover, for
example, the crossover point in the first parent is selected randomly, as in
standard crossover. Then the size of the subtree to be excised is calculated.
This is used to constrain the choice of the second crossover point so as
to guarantee that the subtree chosen from the second parent will not be
unfairly big.
Older methods include several mutation operators that may help control
the average tree size in the population while still introducing new genetic
material. Kinnear (1993) proposes a mutation operator which prevents the
offsprings depth being more than 15% larger than its parent. Langdon
(1998) proposes two mutation operators in which the new random subtree is

106

11 GP Theory and its Applications

on average the same size as the code it replaces. In Hoist mutation (Kinnear,
1994a) the new subtree is selected from the subtree being removed from the
parent, guaranteeing that the new program will be smaller than its parent.
Shrink mutation (Angeline, 1996) is a special case of subtree mutation where
the randomly chosen subtree is replaced by a randomly chosen terminal.
McPhee and Poli (2002) provides theoretical analysis and empirical evidence
that combinations of subtree crossover and subtree mutation operators can
control bloat in linear GP systems.
Other methods which control bloat by exploiting the bias of the operators
were discussed in Section 9.4.
Anti-Bloat Selection
As clarified by the size evolution equation discussed in the previous section,
in systems with symmetric operators, bloat can only happen if there are
some longer-than-average programs that are fitter than average or some
shorter-than-average programs that are less fit than average, or both. So,
it stands to reason that in order to control bloat one needs to somehow
modulate the selection probabilities of programs based on their size.
As we have discussed in Section 9.2.1, recent methods also include the
use of multi-objective optimisation to control bloat. This typically involves
the use of a modified selection based on the Pareto criterion.
A recent technique, the Tarpeian method (Poli, 2003), controls bloat
by acting directly on the selection probabilities in Equation (11.2). This is
done by setting the fitness of randomly chosen longer-than-average programs
to 0. This prevents them being parents. By changing how frequently this
is done the anti-bloat intensity of Tarpeian control can be modulated. An
advantage of the method is that the programs whose fitness is zeroed are
never executed, thereby speeding up runs.
The well-known parsimony pressure method (Koza, 1992; Zhang and
M
uhlenbein, 1993, 1995; Zhang et al., 1997) changes the selection probabilities by subtracting a value based on the size of each program from its fitness.
Bigger programs have more subtracted and, so, have lower fitness and tend
to have fewer children. That is, the new fitness function is f (x) c `(x),
where `(x) is the size of program x, f (x) is its original fitness and c is a constant known as the parsimony coefficient.2 Zhang and M
uhlenbein (1995)
showed some benefits of adaptively adjusting the coefficient c at each generation but most implementations actually keep the parsimony coefficient
constant.

2 While the new fitness is used to guide evolution, one still needs to use the original
fitness function to recognise solutions and stop runs.

11.3 Bloat

107

The parsimony pressure method can be seen as a way to address the

generalisationaccuracy tradeoff common in machine learning (Rosca and
Ballard, 1996b; Zhang and M
uhlenbein, 1995). There are also connections
between this method and the Minimum Description Length (MDL) principle
used to control bloat in (Iba, 1997; Iba et al., 1994; Iba, de Garis, and
Sato, 1995a). The MDL approach uses a fitness function which combines
program complexity (expressed as the number of bits necessary to encode
the programs tree) and classification error (expressed as the number of
bits necessary to encode the errors on all fitness cases). Rosca also linked
the parsimony pressure method to his approximate evolution equations for
rooted-tree schemata (Rosca, 1996, 1997; Rosca and Ballard, 1996b, 1999).
Controlling bloat while at the same time maximising fitness turns the
evolution of programs into either a multi-objective optimisation problem or,
at least, into a constrained optimisation problem. The parsimony pressure
method effectively treats the minimisation of size as a soft constraint and
attempts to enforce this constraint using the penalty method, i.e., by decreasing the fitness of programs by an amount that depends on their size. The
penalty is typically simply proportional to program size. The intensity with
which bloat is controlled is, therefore, determined by the parsimony coefficient. The value of this coefficient is very important: too small a value and
runs will still bloat wildly; too large a value and GP will take the minimisation of size as its main target and will almost ignore fitness, thus converging
towards extremely small but useless programs (Soule, 1998). However, good
values of the parsimony coefficient are highly dependent on particulars such
as the problem being solved, the choice of functions and terminals, and various parameter settings. Furthermore, with a constant parsimony coefficient
the method can only achieve partial control over the dynamics of the average
program size over time.
Recently, a theoretically sound method for setting the parsimony coefficient in a principled manner has been proposed (Poli and McPhee, 2008b).
The covariant parsimony pressure method is based on an analysis of the size
evolution Equation (11.1), and is easy to implement. It recalculates the parsimony coefficient c at each generation using c = Cov(`, f )/ Var(`), where
Cov(`, f ) is the covariance between program size ` and program fitness f
in the population, and Var(`) is the variance of program sizes. Note that c
needs to be recalculated each generation because both Cov(`, f ) and Var(`)
change from generation to generation. As shown in Figure 11.2 (in the portion labelled Local), using this equation ensures that the mean program
size remains at the value set by the initialisation procedure (although there
can be a small amount of drift). There is a variant of the method that allows
the user to even decide what function the mean program size should follow
over time. As shown in the figure this provides complete control over the
population size dynamics.

108

11 GP Theory and its Applications

Figure 11.2: Plots of the evolution average size over 500 generations for
multiple runs of the 6-MUX problem with various forms of covariant parsimony pressure. The Constant runs had a constant target size of 150. In
the Sin runs the target size was sin((generation + 1)/50.0) 50.0 + 150.
For the Linear runs the target size was 150 + generation. The Limited
runs used no size control until the size reached 250, then the target was held
at 250. Finally, the Local runs used c = Cov(`, f )/ Var(`), which allowed
a certain amount of drift but still avoided runaway bloat (see text).

Part III

Practical Genetic
Programming

Three little pigs provide a demonstration of construction techniques. . .

and Goldilocks finally gets it just right.

109

Chapter 12

Applications
Since its early beginnings, GP has produced a cornucopia of results. The
literature, which covers more than 5000 recorded uses of GP, reports an
enormous number of applications where GP has been successfully used as
an automatic programming tool, a machine learning tool or an automatic
problem-solving engine. It is impossible to list all such applications here.
In the following sections we start with a discussion of the general kinds
of problems where GP has proved successful (Section 12.1) and then review a representative subset for each of the main application areas of GP
(Sections 12.212.11), devoting particular attention to the important areas
of symbolic regression (Section 12.2) and human-competitive results (Section 12.3).

12.1

Where GP has Done Well

Based on the experience of numerous researchers over many years, it appears

that GP and other evolutionary computation methods have been especially
productive in areas having some or all of the following properties:
The interrelationships among the relevant variables is unknown
or poorly understood (or where it is suspected that the current understanding may possibly be wrong). One of the particular values of GP (and other evolutionary algorithms) is in exploring
poorly understood domains. If the problem domain is well understood,
there may well be analytical tools that will provide quality solutions
without the uncertainty inherent in a stochastic search process such
as GP. GP, on the other hand, has proved successful where the application is new or otherwise not well understood. It can help discover
which variables and operations are important; provide novel solutions
111

112

12 Applications
to individual problems; unveil unexpected relationships among variables; and, sometimes GP can discover new concepts that can then be
applied in a wide variety of circumstances.

Finding the size and shape of the ultimate solution is a major

part of the problem. If the form of the solution is known, then
alternative search mechanisms that work on fixed size representations
(e.g., genetic algorithms) may be more efficient because they wont
have to discover the size and shape of the solution.
Significant amounts of test data are available in computerreadable form. GP (and most other machine learning and search
techniques) benefit from having significant pools of test data. At a
minimum there needs to be enough data to allow the system to learn
the salient features, while leaving enough at the end to use for validation and over-fitting tests. It is also useful if the test data are as clean
and accurate as possible. GP is capable of dealing gracefully with
certain amounts of noise in the data (especially if steps are taken to
reduce over-fitting), but cleaner data make the learning process easier
for any system, GP included.
There are good simulators to test the performance of tentative
solutions to a problem, but poor methods to directly obtain
good solutions. In many domains of science and engineering, simulators and analysis tools have been constructed that allow one to
evaluate the behaviour and performance of complex artifacts such as
aircraft, antennas, electronic circuits, control systems, optical systems,
games, etc. These simulators contain enormous amounts of knowledge
of the domain and have often required several years to create. These
tools solve the so-called direct problem of working out the behaviour
of a solution or tentative solution to a problem, given the solution itself. However, the knowledge stored in such systems cannot be easily
used to solve the inverse problem of designing an artifact from a set
of functional or performance requirements. A great advantage of GP
is that it is able to connect to simulators and analysis tools and to
data-mine the simulator to solve the inverse problem automatically.
That is, the user need not specify (or know) much about the form of
the eventual solution before starting.
Conventional mathematical analysis does not, or cannot, provide
analytic solutions. If there is a good exact analytic solution, one
probably wants to use it rather than spend the energy to evolve what
is likely to be an approximate solution. That said, GP might still be
a valuable option if the analytic solutions have undesirable properties
(e.g., unacceptable run times for large instances), or are based on

12.2 Curve Fitting, Data Modelling and Symbolic Regression

113

assumptions that dont apply in ones circumstances (e.g., noise-free

data).
An approximate solution is acceptable (or is the only result that
is ever likely to be obtained). Evolution in general, and GP in
particular, is typically about being good enough rather than the
best. (A rabbit doesnt have to be the fastest animal in the world:
it just has to be fast enough to escape that particular fox.) As a
result, evolutionary algorithms tend to work best in domains where
close approximations are both possible and acceptable.
Small improvements in performance are routinely measured (or
easily measurable) and highly prized. Technological efforts tend
to concentrate in areas of high economic importance. In these domains,
the state of the art tends to be fairly advanced, and, so, it is difficult
to improve over existing solutions. However, in these same domains
small improvements can be extremely valuable. GP can sometimes
discover small, but valuable, relationships.
Two (of many) examples of successful applications of GP that satisfy
many of these properties are the work of Lohn, Hornby, and Linden (2004) on
satellite antenna design and Spectors evolution of new quantum computing
algorithms that out-performed all previous approaches (Spector, Barnum,
and Bernstein, 1998; Spector, Barnum, Bernstein, and Swamy, 1999). Both
of these domains are complex, without analytic solutions, yet in both cases
good simulators existed which could be used to evaluate the fitness of solutions. In other words, people didnt know how to solve the problems but
they could (automatically) recognise a good solution when they saw one.
Both of these applications resulted in the discovery of highly successful and
unexpected designs. The key component of the evolved quantum algorithm
could in fact be extracted and applied in a wide variety of other settings,
leading to major improvements in a number of related quantum algorithms
as well as the ones under specific study.

12.2

Curve Fitting, Data Modelling and

Symbolic Regression

In principle, there are as many possible applications of GP as there are applications for programsin other words, virtually infinite. However, before
one can try to solve a new problem with GP, one needs to define an appropriate fitness function. In problems where only the side effects of a program are
of interest, the fitness function usually compares the effects of the execution
of a program in some suitable environments with a desired behaviour, often
in a very application-dependent manner. However, in many problems the

114

12 Applications

goal is to find a function whose output has some desired property, e.g., the
function matches some target values (as in the example given in Section 4.1).
This is generally known as a symbolic regression problem.
Many people are familiar with the notion of regression. Regression means
finding the coefficients of a predefined function such that the function best
fits some data. A problem with regression analysis is that, if the fit is not
good, the experimenter has to keep trying different functions by hand until
a good model for the data is found. Not only is this laborious, but also
the results of the analysis depend very much on the skills and inventiveness
of the experimenter. Furthermore, even expert users tend to have strong
mental biases when choosing functions to fit. For example, in many application areas there is a considerable tradition of using only linear or quadratic
models, even when the data might be better fit by a more complex model.
Symbolic regression attempts to go beyond this. It consists of finding
a function that fits the given data points without making any assumptions
about the structure of that function. Since GP makes no such assumption,
it is well suited to this sort of discovery task. Symbolic regression was one
of the earliest applications of GP (Koza, 1992), and continues to be widely
studied (Cai, Pacheco-Vega, Sen, and Yang, 2006; Gustafson, Burke, and
Krasnogor, 2005; Keijzer, 2004; Lew, Spencer, Scarpa, Worden, Rutherford,
and Hemez, 2006).
The steps necessary to solve symbolic regression problems include the five
preparatory steps mentioned in Chapter 2. We practiced them in the example in Chapter 4, which was an instance of a symbolic regression problem.
There is an important difference here, however: the data points provided in
Chapter 4 were computed using a simple formula, while in most realistic situations each point represents the measured values taken by some variables
at a certain time in some dynamic process, in a repetition of an experiment,
and so on. So, the collection of an appropriate set of data points for symbolic
regression is an important and sometimes complex task.
For instance, consider the case of using GP to evolve a soft sensor (Jordaan, Kordon, Chiang, and Smits, 2004). The intent is to evolve a function
that will provide a reasonable estimate of what a sensor (in an industrial
production facility) would report, based on data from other actual sensors
in the system. This is typically done in cases where placing an actual sensor
in that location would be difficult or expensive. However, it is necessary to
place at least one instance of such a sensor in a working system in order to
collect the data needed to train and test the GP system. Once the sensor
is placed, one would collect the values reported by that sensor and by all
the other real sensors that are available to the evolved function, at various
times, covering the various conditions under which the evolved system will
be expected to act.
Such experimental data typically come in large tables where numerous

12.2 Curve Fitting, Data Modelling and Symbolic Regression

115

quantities are reported. Usually we know which variable we want to predict

(e.g., the soft sensor value), and which other quantities we can use to make
the prediction (e.g., the hard sensor values). If this is not known, then
experimenters must decide which are going to be their dependent variables
before applying GP. Sometimes, in practical situations, the data tables
include hundreds or even thousands of variables. It is well known that
in these cases the efficiency and effectiveness of any machine learning or
program induction method, including GP, can dramatically drop as most of
the variables are typically redundant or irrelevant. This forces the system
to waste considerable energy on isolating the key features. To avoid this,
it is necessary to perform some form of feature selection, i.e., we need to
decide which independent variables to keep and which to leave out. There
are many techniques to do this, but these are beyond the scope of this book.
However, it is worth noting that GP itself can be used to do feature selection
as shown in (Langdon and Buxton, 2004).
There are problems where more than one output (prediction) is required.
For example, Table 12.1 shows a data set with four variables controlled
during data collection (left) and six dependent variables (right). The data
were collected for the purpose of solving an inverse kinematics problem in the
Elvis robot (Langdon and Nordin, 2001). The robot is shown in Figure 12.1
during the acquisition of a data sample. The roles of the independent and
dependent variables are swapped when GP is given the task of controlling
the arm given data from the robots eyes.
There are several GP techniques which might be used to deal with applications where multiple outputs are required: GP individuals including
multiple trees (as in Figure 2.2, page 11), linear GP with multiple output
registers (see Section 7.1), graph-based GP with multiple output nodes (see
Section 7.2), a single GP tree with primitives operating on vectors, and so
forth.
Once a suitable data set is available, its independent variables must all
be represented in the primitive set. What other terminals and functions are
included depends very much on the type of the data being processed (are
they numeric? are they strings? etc.) and is often guided by the information
available to the experimenter and the process that generated the data. If
something is known (or strongly suspected) about the desired structure of
the function to be evolved, it may be very beneficial to use this information
(or to apply some constraints, like those discussed in Section 6.2). For
example, if the data are known to be periodic, then the function set should
probably include something like the sine function.
What is common to virtually all symbolic regression problems is that
the fitness function must measure how close the outputs produced by each
program are to the values of the dependent variables, when the corresponding values of the independent ones are used as inputs for the program. So,

116

12 Applications

Table 12.1: Samples showing the size and location of Elviss finger tip
as apparent to this two eyes, given various right arm actuator set points (4
degrees of freedom). Cf. Figure 12.1. When the data are used for training,
GP is asked to invert the mapping and evolve functions from data collected
by both cameras showing a target location to instructions to give to Elviss
four arm motors so that its arm moves to the target.
Arm actuator
-376
-372
-377
-385
-393
-401
-409
-417
-425
-433
-441
..
.

-626
-622
-627
-635
-643
-651
-659
-667
-675
-683
-691
..
.

1000
1000
899
799
699
599
500
399
299
199
99
..
.

-360
-380
-359
-319
-279
-239
-200
-159
-119
-79
-39
..
.

x
44
43
43
38
36
32
32
31
30
31
31
..
.

Left eye
y size
10
29
7
29
9
33
16
27
24
26
32
25
35
24
41
17
45
25
47
20
49
16
..
..
.
.

Right eye
x
y size
-9 12
25
-9 12
29
-20 14
26
-17 22
30
-21 25
20
-26 28
18
-27 31
19
-28 36
13
-27 39
8
-27 43
9
-26 45
13
..
..
..
.
.
.

continues for a total of 691 lines

most symbolic regression fitness functions tend to include summing the errors measured for each record in the data set, as we did in Section 4.2.2.
Usually either the absolute difference or the square of the error is used.

The fourth preparatory step typically involves choosing a size for the
population (which is often done initially based on the perceived difficulty of
the problem, and is then refined based on the actual results of preliminary
runs). The user also needs to set the balance between the selection strength
(normally tuned via the tournament size) and the intensity of variation
(which can be varied by modifying the mutation and crossover rates, but
many researchers tend to fix to some standard values).

12.3 Human Competitive Results the Humies

117

Figure 12.1: Elvis sitting with its right hand outstretched. The apparent
position and size of a bright red laser attached to its finger tip is recorded
(see Table 12.1). The data are then used to train a GP to move the robots
arm to a spot in three dimensions using only its eyes.

12.3

Human Competitive Results:

The Humies

Getting machines to produce human-like results is the very reason for the
existence of the fields of artificial intelligence and machine learning. However, it has always been very difficult to assess how much progress these
fields have made towards their ultimate goal. Alan Turing understood that
in order to avoid human biases when assessing machine intelligence, machine
behaviour must be evaluated objectively. This led him to propose an imitation game, now known as the Turing test (Turing, 1950). Unfortunately,
the Turing test is not usable in practice, and so, there is a need for more
workable objective tests of machine intelligence.
Koza, Bennett, and Stiffelman (1999) suggested shifting attention from
the notion of intelligence to the notion of human competitiveness. A result
cannot acquire the rating of human competitive merely because it is endorsed by researchers inside the specialised fields that are attempting to
create machine intelligence. A result produced by an automated method

118

12 Applications

must earn the rating of human competitive independently of the fact that
it was generated by an automated method.
Koza proposed that an automatically-created result should be considered
human-competitive if it satisfies at least one of these eight criteria:
1. The result was patented as an invention in the past, is an improvement
over a patented invention or would qualify today as a patentable new
invention.
2. The result is equal to or better than a result that was accepted as a new
scientific result at the time when it was published in a peer-reviewed
scientific journal.
3. The result is equal to or better than a result that was placed into a
database or archive of results maintained by an internationally recognised panel of scientific experts.
4. The result is publishable in its own right as a new scientific result,
independent of the fact that the result was mechanically created.
5. The result is equal to or better than the most recent human-created
solution to a long-standing problem for which there has been a succession of increasingly better human-created solutions.
6. The result is equal to or better than a result that was considered an
achievement in its field at the time it was first discovered.
7. The result solves a problem of indisputable difficulty in its field.
8. The result holds its own or wins a regulated competition involving
human contestants (in the form of either live human players or humanwritten computer programs).
These criteria are independent of, and at arms length from, the fields of
artificial intelligence, machine learning, and GP.
Over the years, dozens of results have passed the human-competitiveness
test. Some pre-2004 human-competitive results include:
Creation of quantum algorithms, including a better-than-classical algorithm for a database search problem and a solution to an AND/OR
query problem (Spector et al., 1998, 1999).
Creation of a competitive soccer-playing program for the RoboCup 1997
competition (Luke, 1998).
Creation of algorithms for the transmembrane segment identification
problem for proteins (Koza, 1994, Sections 18.8 and 18.10) and (Koza
et al., 1999, Sections 16.5 and 17.2).

12.3 Human Competitive Results the Humies

119

Creation of a sorting network for seven items using only 16 steps (Koza
et al., 1999, Sections 21.4.4, 23.6, and 57.8.1).
Synthesis of analogue circuits (with placement and routing, in some
cases), including: 60- and 96-decibel amplifiers (Koza et al., 1999,
Section 45.3); circuits for squaring, cubing, square root, cube root,
logarithm, and Gaussian functions (Koza et al., 1999, Section 47.5.3);
a circuit for time-optimal control of a robot (Koza et al., 1999, Section
48.3); an electronic thermometer (Koza et al., 1999, Section 49.3); a
voltage-current conversion circuit (Koza, Keane, Streeter, Mydlowec,
Yu, and Lanza, 2003, Section 15.4.4).
Creation of a cellular automaton rule for the majority classification
problem that is better than all known rules written by humans (Andre
et al., 1996).
Synthesis of topology for controllers, including: a PID (proportional,
integrative, and derivative) controller (Koza et al., 2003, Section 9.2)
and a PID-D2 (proportional, integrative, derivative, and second derivative) controller (Koza et al., 2003, Section 3.7); PID tuning rules that
outperform the Ziegler-Nichols and Astrom-Hagglund tuning rules
(Koza et al., 2003, Chapter 12); three non-PID controllers that outperform a PID controller that uses the Ziegler-Nichols or AstromHagglund tuning rules (Koza et al., 2003, Chapter 13).
In total (Koza and Poli, 2005) lists 36 human-competitive results. These
include 23 cases where GP has duplicated the functionality of a previously
patented invention, infringed a previously patented invention, or created a
patentable new invention. Specifically, there are fifteen examples where GP
has created an entity that either infringes or duplicates the functionality of
a previously patented 20th -century invention, six instances where GP has
done the same with respect to an invention patented after 1 January 2000,
and two cases where GP has created a patentable new invention. The two
new inventions are general-purpose controllers that outperform controllers
employing tuning rules that have been in widespread use in industry for
most of the 20th century.
Many of the pre-2004 results were obtained by Koza. However, since
2004, a competition has been held annually at ACMs Genetic and Evolutionary Computation Conference (termed the Human-Competitive awards
the Humies). The $10,000 prize is awarded to projects that have produced automatically-created results which equal or better those produced
by humans.
The Humies Prizes have typically been awarded to applications of evolutionary computation to high-tech fields. Many used GP. For example,
the 2004 gold medals were given for the design, via GP, of an antenna for

120

12 Applications

Figure 12.2: Award winning human-competitive antenna design produced

by GP.

deployment on NASAs Space Technology 5 Mission (see Figure 12.2) (Lohn

et al., 2004) and for evolutionary quantum computer programming (Spector, 2004). There were three silver medals in 2004: one for the evolution of
local search heuristics for SAT using GP (Fukunaga, 2004), one for the application of GP to the synthesis of complex kinematic mechanisms (Lipson,
2004) and one for organisation design optimisation using GP (KHosraviani,
2003; KHosraviani, Levitt, and Koza, 2004). Also, four of the 2005 medals
were awarded for GP applications: the invention of optical lens systems (AlSakran, Koza, and Jones, 2005; Koza, Al-Sakran, and Jones, 2005), the evolution of a quantum Fourier transform algorithm (Massey, Clark, and Stepney, 2005), evolving assembly programs for Core War (Corno, Sanchez, and
Squillero, 2005) and various high-performance game players for Backgammon, Robocode and Chess endgame (Azaria and Sipper, 2005a,b; Hauptman and Sipper, 2005; Shichel, Ziserman, and Sipper, 2005). In 2006, GP
again scored a gold medal with the synthesis of interest point detectors for
image analysis (Trujillo and Olague, 2006a,b), while it scored a silver medal
in 2007 with the evolution of an efficient search algorithm for the Mate-in-N
problem in Chess (Hauptman and Sipper, 2007) (see Figure 12.3).
Note that many human competitive results were presented at the Humies
20042007 competitions (e.g., 11 of the 2004 entries were judged to be human
competitive). However, only the very best were awarded medals. So, at the

12.4 Image and Signal Processing

121

Figure 12.3: Example mate-in-2 problem.

time of writing we estimate that there are at least 60 human competitive

results obtained by GP. This shows GPs potential as a powerful invention
machine.

12.4

Image and Signal Processing

Hampo and Marko (1992) were among the first people from industry to
consider using GP for signal processing. They evolved algorithms for preprocessing electronic motor vehicle signals for possible use in engine monitoring and control.
Several applications of GP for image processing have been for military
uses. For example, Tackett (1993) evolved algorithms to find tanks in infrared images. Howard, Roberts, and Brankin (1999); Howard, Roberts, and
Ryan (2006) evolved programs to pick out ships from SAR radar mounted
on satellites in space and to locate ground vehicles from airborne photo reconnaissance. They also used GP to process surveillance data for civilian
purposes, such as predicting motorway traffic jams from subsurface traffic
speed measurements (Howard and Roberts, 2004).
Using satellite SAR radar, Daida, Hommes, Bersano-Begey, Ross, and
Vesecky (1996) evolved algorithms to find features in polar sea ice. Optical satellite images can also be used for environmental studies (Chami and
Robilliard, 2002) and for prospecting for valuable minerals (Ross, Gualtieri,
Fueten, and Budkewitsch, 2005).
Alcazar used GP to find recurrent filters (including artificial neural networks (Esparcia-Alcazar and Sharman, 1996)) for one-dimensional electronic

122

12 Applications

signals (Sharman and Esparcia-Alcazar, 1993). Local search (simulated annealing or gradient descent) can be used to adjust or fine-tune constant
values within the structure created by genetic search (Smart and Zhang,
2004).
Yu and Bhanu (2006) have used GP to preprocess images, particularly
of human faces, to find regions of interest for subsequent analysis. See also
(Trujillo and Olague, 2006a).
Zhang has been particularly active at evolving programs with GP to
visually classify objects (typically coins) (Zhang and Smart, 2006). He has
also applied GP to human speech (Xie, Zhang, and Andreae, 2006).
Parisian GP is a system in which the image processing task is split
across a swarm of evolving agents (flies). In (Louchet, 2001; Louchet,
Guyon, Lesot, and Boumaza, 2002) the flies reconstruct three dimensions
from pairs of stereo images. For example, in (Louchet, 2001), as the flies
buzz around in three dimensions their position is projected onto the left and
right of a pair of stereo images. The fitness function tries to minimise the
discrepancy between the two images, thus encouraging the flies to settle on
visible surfaces in the 3-D space. So, the true 3-D space is inferred from
pairs of 2-D images taken from slightly different positions.
While the likes of Google have effectively indexed the written word, for
speech and pictures indexing has been much less effective. One area where
GP might be applied is in the automatic indexing of images. Some initial
steps in this direction are given in (Theiler, Harvey, Brumby, Szymanski,
Alferink, Perkins, Porter, and Bloch, 1999).
To some extent, extracting text from images (OCR) can be done fairly
reliably, and the accuracy rate on well formed letters and digits is close
to 100%. However, many interesting cases remain (Cilibrasi and Vitanyi,
2005) such as Arabic (Klassen and Heywood, 2002) and oriental languages,
handwriting (De Stefano, Cioppa, and Marcelli, 2002; Gagne and Parizeau,
2006; Krawiec, 2004; Teredesai and Govindaraju, 2005) (such as the MNIST
examples), other texts (Rivero, nal, Dorado, and Pazos, 2004) and musical
scores (Quintana, Poli, and Claridge, 2006).
The scope for applications of GP to image and signal processing is almost
unbounded. A promising area is medical imaging (Poli, 1996b). GP image
techniques can also be used with sonar signals (Martin, 2006). Off-line work
on images includes security and verification. For example, Usman, Khan,
Chamlawi, and Majid (2007) have used GP to detect image watermarks
which have been tampered with. Recent work by Zhang has incorporated
multi-objective fitness into GP image processing (Zhang and Rockett, 2006).
In 1999 Poli, Cagnoni and others founded the annual European Workshop on Evolutionary Computation in Image Analysis and Signal Processing
(EvoIASP). EvoIASP is held every year with the EuroGP. Whilst not solely
dedicated to GP, many GP applications have been presented at EvoIASP.

12.5 Financial Trading, Time Series, and Economic Modelling

12.5

123

Financial Trading, Time Series

Prediction and Economic Modelling

GP is very widely used in the areas of financial trading, time series prediction
and economic modelling and it is impossible to describe all its applications.
It this section we will hint at just a few areas.
Chen has written more than 60 papers on using GP in finance and economics. Recent papers have looked at the modelling of agents in stock
markets (Chen and Liao, 2005), game theory (Chen, Duffy, and Yeh, 2002),
evolving trading rules for the S&P 500 (Yu and Chen, 2004) and forecasting
the Heng-Sheng index (Chen, Wang, and Zhang, 1999).
The efficient markets hypothesis is a tenet of economics. It is founded
on the idea that everyone in a market has perfect information and acts
rationally. If the efficient markets hypothesis held, then everyone would
see the same value for items in the market and so agree the same price.
Without price differentials, there would be no money to be made from the
market itself. Whether it is trading potatoes in northern France or dollars
for yen, it is clear that traders are not all equal and considerable doubt has
been cast on the efficient markets hypothesis. So, people continue to play the
stock market. Game theory has been a standard tool used by economists to
try to understand markets but is increasingly supplemented by simulations
with both human and computerised agents. GP is increasingly being used
as part of these simulations of social systems.
Neely, Weller, and Dittmar (1997), Neely and Weller (1999, 2001) and
Neely (2003) of the US Federal Reserve Bank used GP to study intra-day
technical trading on the foreign exchange markets to suggest the market is
efficient and found no evidence of excess returns. This negative result
was criticised by Marney, Miller, Fyfe, and Tarbert (2001). Later work by
Neely, Weller, and Ulrich (2006) suggested that data after 1995 are consistent with Los adaptive markets hypothesis rather than the efficient markets
hypothesis. Note that here GP and computer tools are being used in a
novel data-driven approach to try and resolve issues which were previously
a matter of dogma.
From a more pragmatic viewpoint, Kaboudan shows GP can forecast international currency exchange rates (Kaboudan, 2005), stocks (Kaboudan,
2000) and stock returns (Kaboudan, 1999). Tsang and his co-workers continue to apply GP to a variety of financial arenas, including: betting (Tsang,
Li, and Butler, 1998), forecasting stock prices (Li and Tsang, 1999; Tsang
and Li, 2002; Tsang, Yung, and Li, 2004), studying markets (MartinezJaramillo and Tsang, 2007), approximating Nash equilibrium in game theory (Jin, 2005; Jin and Tsang, 2006; Tsang and Jin, 2006) and arbitrage
(Tsang, Markose, and Er, 2005). Dempster and HSBC also use GP in foreign exchange trading (Austin, Bates, Dempster, Leemans, and Williams,

124

12 Applications

2004; Dempster and Jones, 2000; Dempster, Payne, Romahi, and Thompson,
2001). Pillay has used GP in social studies and teaching aids in education,
e.g. (Pillay, 2003). As well as trees (Koza, 1990), other types of GP have
been used in finance, e.g. (Nikolaev and Iba, 2002).
Since 1995 the International Conference on Computing in Economics
and Finance (CEF) has been held every year. It regularly attracts GP papers, many of which are on-line. In 2007 Brabazon and ONeill established
the European Workshop on Evolutionary Computation in Finance and Economics (EvoFIN). EvoFIN is held with EuroGP.

12.6

Industrial Process Control

There is evidence that GP is frequently used in industrial process control,

although, of course, most industrialists have little time to spend on academic reporting. A notable exception is Dow Chemical, where a group has
been very active in publishing results (Castillo, Kordon, and Smits, 2006a;
Castillo, Kordon, Smits, Christenson, and Dickerson, 2006b; Jordaan, den
Doelder, and Smits, 2006; Kordon, Castillo, Smits, and Kotanchek, 2005;
Kotanchek et al., 2006; Mercure, Smits, and Kordon, 2001). Kordon (2006)
describes where industrial GP stands now and how it will progress.
Another active collaboration is that of Kovacic and Balic, who used GP
in the computer numerical control of industrial milling and cutting machinery (Kovacic and Balic, 2003). The partnership of Deschaine and Francone
(Francone and Deschaine, 2004) is most famous for their use of Discipulus (Foster, 2001) for detecting bomb fragments and unexploded ordnance
(Deschaine, 2006). Discipulus has also been used as an aid in the development of control systems for rubbish incinerators (Deschaine, Patel, Guthrie,
Grimski, and Ades, 2001).
One of the earliest users of GP in control was Willis Chemical Engineering group at Newcastle, which used GP to model flow in a plasticating
extruder (Willis, Hiden, and Montague, 1997a). Other GP applications in
the plastics industry include (Brezocnik, Balic, and Gusel, 2000). McKay,
Willis, Searson, and Montague (2000) also modelled extruding food. Searson, Montague, and Willis (1998) modelled control of chemical reactions in
continuous stirred tank reactors. Marenbach (1998) investigated GP in the
control of biotech reactors. Willis, Hiden, Marenbach, McKay, and Montague (1997b) surveyed GP applications, including in the area of control.
Lewin, Lachman-Shalem, and Grosman (2006) and Dassau, Grosman,
and Lewin (2006) applied GP to the control of an integrated circuit fabrication plant. Domingos worked on simulations of nuclear reactors (PWRs to
be exact) to devise better ways of preventing xenon oscillations (Domingos,
Schirru, and Martinez, 2005). GP has also been used to identify the state
of a plant to be controlled (in order to decide which of various alternative

12.7 Medicine, Biology and Bioinformatics

125

control laws to apply). For example, Flemings group in Sheffield used multiobjective GP (Hinchliffe and Willis, 2003; Rodriguez-Vazquez, Fonseca, and
Fleming, 2004) to reduce the cost of running aircraft jet engines (Arkov,
Evans, Fleming, Hill, Norton, Pratt, Rees, and Rodriguez-Vazquez, 2000;
Evans, Fleming, Hill, Norton, Pratt, Rees, and Rodriguez-Vazquez, 2001).
Alves da Silva and Abrao (2002) surveyed GP and other AI techniques
applied in the electrical power industry.

12.7

Medicine, Biology and Bioinformatics

GP has long been applied to medicine, biology and bioinformatics. Early

work by Handley (1993) and Koza and Andre (1996) used GP to make predictions about the behaviour and properties of biological systems, principally
proteins. Oakley, a practising medical doctor, used GP to model blood flow
in toes (Oakley, 1994) as part of his long term interests in frostbite.
In 2002 Banzhaf and Foster organised BioGEC: the first GECCO workshop on biological applications of genetic and evolutionary computation.
BioGEC has become a bi-annual feature of the annual GECCO conference.
Half a year later Marchiori and Corne organised EvoBio: the European conference on evolutionary computation, machine learning and data mining in
bioinformatics. EvoBio is held every year alongside EuroGP. GP figures
heavily in both BioGEC and EvoBIO.
GP is often used in biomedical data mining. Of particular medical interest are very wide data sets, with many inputs per sample (Lavington,
Dewhurst, Wilkins, and Freitas, 1999). Examples include infrared spectra
(Ellis, Broadhurst, and Goodacre, 2004; Ellis, Broadhurst, Kell, Rowland,
and Goodacre, 2002; Goodacre, 2003; Goodacre, Shann, Gilbert, Timmins,
McGovern, Alsberg, Kell, and Logan, 2000; Harrigan, LaPlante, Cosma,
Cockerell, Goodacre, Maddox, Luyendyk, Ganey, and Roth, 2004; Johnson, Gilbert, Winson, Goodacre, Smith, Rowland, Hall, and Kell, 2000;
McGovern, Broadhurst, Taylor, Kaderbhai, Winson, Small, Rowland, Kell,
and Goodacre, 2002; Taylor, Goodacre, Wade, Rowland, and Kell, 1998;
Vaidyanathan, Broadhurst, Kell, and Goodacre, 2003), single nuclear polymorphisms (Barrett, 2003; Reif, White, and Moore, 2004; Shah and Kusiak,
2004), chest pain (Bojarczuk, Lopes, and Freitas, 2000), and Affymetrix
GeneChip microarray data (de Sousa, de C. T. Gomes, Bezerra, de Castro,
and Von Zuben, 2004; Eriksson and Olsson, 2004; Heidema, Boer, Nagelkerke, Mariman, van der A, and Feskens, 2006; Ho, Hsieh, Chen, and Huang,
2006; Hong and Cho, 2006; Langdon and Buxton, 2004; Li, Jiang, Li, Moser,
Guo, Du, Wang, Topol, Wang, and Rao, 2005; Linden and Bhaya, 2007; Yu,
Yu, Almal, Dhanasekaran, Ghosh, Worzel, and Chinnaiyan, 2007).

126

12 Applications

Kell and his colleagues in Aberystwyth have had great success in applying
GP widely in bioinformatics (see infrared spectra above and (Allen, Davey,
Broadhurst, Heald, Rowland, Oliver, and Kell, 2003; Day, Kell, and Griffith,
2002; Gilbert, Goodacre, Woodward, and Kell, 1997; Goodacre and Gilbert,
1999; Jones, Young, Taylor, Kell, and Rowland, 1998; Kell, 2002a,b,c; Kell,
Darby, and Draper, 2001; Shaw, Winson, Woodward, McGovern, Davey,
Kaderbhai, Broadhurst, Gilbert, Taylor, Timmins, Goodacre, Kell, Alsberg,
and Rowland, 2000; Woodward, Gilbert, and Kell, 1999)). Another very
active group is that of Moore and his colleagues (Moore, Parker, Olsen, and
Aune, 2002; Motsinger, Lee, Mellick, and Ritchie, 2006; Ritchie, Motsinger,
Bush, Coffey, and Moore, 2007; Ritchie, White, Parker, Hahn, and Moore,
2003).
Computational chemistry is widely used in the drug industry. The properties of simple molecules can be calculated. However, the interactions between chemicals which might be used as drugs and medicinal targets within
the body are beyond exact calculation. Therefore, there is great interest in
the pharmaceutical industry in approximate in silico models which attempt
to predict either favourable or adverse interactions between proto-drugs and
biochemical molecules. Since these are computational models, they can be
applied very cheaply in advance of the manufacturing of chemicals, to decide
which of the myriad of chemicals might be worth further study. Potentially,
such models can make a huge impact both in terms of money and time
without being anywhere near 100% correct. Machine learning and GP have
both been tried. GP approaches include (Bains, Gilbert, Sviridenko, Gascon, Scoffin, Birchall, Harvey, and Caldwell, 2002; Barrett and Langdon,
2006; Buxton, Langdon, and Barrett, 2001; Felton, 2000; Globus, Lawton,
and Wipke, 1998; Goodacre, Vaidyanathan, Dunn, Harrigan, and Kell, 2004;
Harrigan et al., 2004; Hasan, Daugelat, Rao, and Schreiber, 2006; Krasnogor, 2004; Si, Wang, Zhang, Hu, and Fan, 2006; Venkatraman, Dalby, and
Yang, 2004; Weaver, 2004).

12.8

GP to Create Searchers and Solvers

Hyper-heuristics

Hyper-heuristics could simply be defined as heuristics to choose other

heuristics (Burke, Kendall, Newall, Hart, Ross, and Schulenburg, 2003).
A heuristic is considered as a rule-of-thumb or educated guess that reduces the search required to find a solution. The difference between metaheuristics and hyper-heuristics is that the former operate directly on the
problem search space with the goal of finding optimal or near-optimal solutions. The latter, instead, operate on the heuristics search space (which
consists of the heuristics used to solve the target problem). The goal then

12.9 Entertainment and Computer Games

127

is finding or generating high-quality heuristics for a problem, for a certain

class of instances of a problem, or even for a particular instance.
GP has been very successfully used as a hyperheuristic. For example, GP
has evolved competitive SAT solvers (Bader-El-Den and Poli, 2007a,b; Fukunaga, 2002; Kibria and Li, 2006), state-of-the-art or better than state-of-theart bin packing algorithms (Burke, Hyde, and Kendall, 2006; Burke, Hyde,
Kendall, and Woodward, 2007; Poli, Woodward, and Burke, 2007), particle
swarm optimisers (Poli, Di Chio, and Langdon, 2005; Poli, Langdon, and
Holland, 2005), evolutionary algorithms (Oltean, 2005), and travelling salesman problem solvers (Keller and Poli, 2007a,b,c; Oltean and Dumitrescu,
2004).

12.9 3/4 Entertainment and Computer Games

Today, a major usage of computers is interactive games (Priesterjahn,
Kramer, Weimer, and Goebels, 2006). There has been some work on incorporating artificial intelligence into mainstream commercial games. The
software owners are not keen on explaining exactly how much AI they use
or giving away sensitive information on how they use AI. Work on GP and
games includes (Azaria and Sipper, 2005a; Langdon and Poli, 2005; Vowk,
Wait, and Schmidt, 2004) as well as the human-competitive game players
mentioned in Section 12.3, page 120. Funes reports experiments which attracted thousands of people via the Internet who were entertained by evolved
Tron players (Funes, Sklar, Juille, and Pollack, 1998b).
Since 2004, the annual IEEE CEC conference has included sessions on
evolutionary computation in games. After chairing the IEEE Symposium
on Computational Intelligence and Games 2005, at Essex University, Simon Lucas founded the IEEE Computational Intelligence Societys Technical Committee on Games. GP features heavily in the Games TCs activities having being applied to Othello, poker, backgammon, draughts, chess,
Ms Pac-Man, robotic football and radio controlled model car racing.

12.10

The Arts

Computers have long been used to create purely aesthetic artifacts. Much
of todays computer art tends to ape traditional drawing and painting, producing static pictures on a computer monitor. However, the immediate
advantage of the computer screen movement can also be exploited. In
both cases evolutionary computation can and has been exploited. Indeed,
with evolutions capacity for unlimited variation, evolutionary computation
offers the artist the scope to produce ever changing works. Some artists
have also worked with sound.

128

12 Applications

The use of GP in computer art can be traced back at least to the work
of Sims (Sims, 1991) and Latham.1 Jacobs work (Jacob, 2000, 2001) provides many examples. McCormack (2006) considers the recent state of play
in evolutionary art and music. Many recent techniques are described in
(Machado and Romero, 2008).
Evolutionary music (Todd and Werner, 1999) has been dominated by
Jazz (Spector and Alpern, 1994). An exception is Bach (Federman, Sparkman, and Watt, 1999). Most approaches to evolving music have made at
least some use of interactive evolution (Takagi, 2001) in which the fitness
of programs is provided by users, often via the Internet (Ando, Dahlsted,
Nordahl, and Iba, 2007; Chao and Forrest, 2003). The limitation is almost always finding enough people willing to participate (Langdon, 2004).
Costelloe and Ryan (2007) tried to reduce the human burden. Algorithmic
approaches are also possible (Cilibrasi, Vitanyi, and de Wolf, 2004; Inagaki,
2002).
One of the sorrows of AI is that as soon as it works it stops being AI (and
celebrated as such) and becomes computer engineering. For example, the
use of computer generated images has recently become cost effective and is
widely used in Hollywood. One of the standard state-of-the-art techniques
is the use of Reynolds swarming boids (Reynolds, 1987) to create animations of large numbers of rapidly moving animals. This was first used in
Cliffhanger (1993) to animate a cloud of bats. Its use is now commonplace
(herds of wildebeest, schooling fish, and even large crowds of people). In
1997 Reynold was awarded an Oscar.
Since 2003, EvoMUSART (the European Workshop on Evolutionary Music and Art) has been held every year along with the EuroGP conference as
part of the EvoStar event.

12.11

Compression

Koza (1992) was the first to use genetic programming to perform compression. He considered, in particular, the lossy compression of images. The idea
was to treat an image as a function of two variables (the row and column
of each pixel) and to use GP to evolve a function that matches as closely as
possible the original. One can then use the evolved GP tree as a lossy compressed version of the image, since it is possible to obtain the original image
by evaluating the program at each row-column pair of interest. The technique, which was termed programmatic compression, was tested on one small
synthetic image with good success. Programmatic compression was further
developed and applied to realistic data (images and sounds) by Nordin and
Banzhaf (1996).
1 http://www.williamlatham1.com/

12.11 Compression

129

Iterated Functions System (IFS) are important in the domain of fractals and the fractal compression algorithm. Lutton, Levy-Vehel, Cretin,
Glevarec, and Roll (1995a,b) used genetic programming to solve the inverse
problem of identifying a mixed IFS whose attractor is a specific binary (black
and white) image of interest. The evolved program can then be taken to represent the original image. In principle, this can then be further compressed.
The technique is lossy, since rarely the inverse problem can be solved exactly. No practical application or compression ratio results were reported
in (Lutton et al., 1995a,b). Using similar principles, Sarafopoulos (1999)
used GP to evolve affine IFSs whose attractors represent a binary image
containing a square (which was compressed exactly) and one containing a
fern (which was achieved with some error in the finer details).
Wavelets are frequently used in lossy image and signal compression.
Klappenecker and May (1995) used GP to evolve wavelet compression algorithms (internal nodes represented conjugate quadrature filters, leaves represented quantisers). Results on a small set of real-world images were impressive, with the GP compression outperforming JPEG at all compression
ratios.
The first lossless compression technique (Fukunaga and Stechert, 1998)
used GP to evolve non-linear predictors for images. These were used to
predict the gray level a pixel will take based on the gray values of a subset
of its neighbours (those that have already been computed in a row-by-row
and column-by-column scan of the image array). The prediction errors together with the models description represent a compressed version of the
image. These were compressed using the Huffman encoding. Results on five
images from the NASA Galileo Mission database were very promising with
GP compression outperforming some of the best human-designed lossless
compression algorithms.
In many compression algorithms some form of pre-processing or transformation of the original data is performed before compression. This often improves compression ratios. Parent and Nowe (2002) evolved pre-processors
for image compression using GP. The objective of the pre-processor was to
reduce losslessly the entropy in the original image. In tests with five images
from the Canterbury corpus, GP was successful in significantly reducing the
image entropy. As verified via the application of bzip2, the resulting images
were markedly easier to compress.
In (Krantz, Lindberg, Thorburn, and Nordin, 2002) the use of programmatic compression was extended from images to natural videos. A program
was evolved that generates intermediate frames of video sequence, where
each frame is composed by a series of transformed regions from the adjacent
frames. The results were encouraging in the sense that a good approximation to frames was achieved. While a significant improvement in compression was achieved, programmatic compression was very slow in comparison

130

12 Applications

with standard compression methods, the time needed for compression being
measured in hours or even days. Acceleration in GP image compression was
achieved in (He, Wang, Zhang, Wang, and Fang, 2005), where an optimal
linear predictive technique was proposed which used a less complex fitness
function.
Recently Kattan and Poli (2008) proposed a GP system called GP-ZIP
for lossless data compression based on the idea of optimally combining wellknown lossless compression algorithms. The file to be compressed was divided into chunks of a predefined length, and GP was asked to find the best
possible compression algorithm for each chunk in such a way as to minimise
the total length of the compressed file. The compression algorithms available to GP-ZIP included arithmetic coding, Lempel-Ziv-Welch, unbounded
prediction by partial matching, and run length encoding among others. Experimentation showed that when the file to be compressed is composed of
heterogeneous data fragments (as it is the case, for example, in archive files),
GP-zip is capable of achieving compression ratios that are significantly superior to those obtained with other compression algorithms.

Chapter 13

Troubleshooting GP
The dynamics of evolutionary algorithms (including GP) are often very complex, and the behaviour of an EA is typically challenging to predict or understand. As a result it is often difficult to troubleshoot such systems when
they are not performing as expected. While we obviously cannot provide
troubleshooting suggestions that are specific to every GP implementation
and application, we can suggest some general issues to keep in mind. To a
large extent the advice in (Kinnear, 1994b; Koza, 1992; Langdon, 1998) also
remains sound.

13.1

Is there a Bug in the Code?

Machine learning systems are notoriously difficult to protect from coding

and logical mistakes. Unless a mistake produces a runtime error, it may
remain hidden in a system for a long time and may contribute to the system
achieving unsatisfactory results. Such mistakes are difficult to find because
the system, being adaptive, will still work to some degree. This is also true
of GP.
The most common reaction to a system not producing satisfactory results
is to start playing with the parameters, the fitness function, the primitive set,
etc. However, one should also consider the possibility of a coding mistake.
The normal program validation techniques, such as inspection of critical
regions of code, should be used to ensure everything is alright.
If the code is part of an established GP implementation, coding errors
are less likely.1 A more probable source of coding errors is stretching the
GP library beyond its original intended use. Reading the manual carefully
is sometimes a good preventive cure for problems.
1 Coding errors cannot be entirely excluded, though, especially if a GP library is large
and provides a rich set of features and functionalities.

131

132

13.2

13 Troubleshooting GP

Can you Trust your Results?

Since GP is a stochastic search algorithm, different runs may have different

outcomes and yield different results. Because of this, one needs to be very
careful in making inferences regarding the degree of success of the system
from a small set of runs.
It is possible, for example, to run a GP system 10 times on a particular
problem, observe that all 10 runs failed to find a solution, and conclude that
GP cannot solve the problem. However, if the success probability is say 5%
with a particular choice of parameters and representation, the probability of
doing 10 runs and all of them failing is almost 60%! So, the failure to solve
the problem in these 10 runs should not come as a surprise, even though
theres a reasonable chance that you would find a solution if you did more
runs.
For precisely this reason, it is very important to do enough runs and
use appropriate statistical tests to ensure that conclusions are statistically
significant.
GP runs can often be very time consuming, especially if the fitness function is computationally expensive. While parallel and distributed computing
(see Section 10.4) can significantly speed up the process, tools from the design of experiments literature (Bartz-Beielstein, 2006) can also be used to
reduce the number of different runs that are necessary to explore the space
in a statistically sound manner.
A common GP application is classification, e.g., evolving a program or
function that can classify patient biopsy data into two categories: cancerous
or benign. There are numerous pitfalls in this type of work, such as using
all the available data as training data, thereby leaving nothing to use for
validating your evolved solution on unseen data. There is a broad literature
on this and related subjects, and numerous tools such as cross-validation
that one can use when not enough data are available. (See, for example,
(Hastie, Tibshirani, and Friedman, 2001).) The aim must be to ensure that
your results can be trusted to work in the real world, rather than in just the
synthetic environment created by the fitness cases we chose.

13.3

There are No Silver Bullets

When working on real problems there are not likely to be any silver bullets.
No technique (including GP) is likely to solve all instances of an NP-hard
problem in an amount of time that grows linearly with the size of the problem. GP has proven extremely successful in a wide variety of domains (e.g.,
Chapter 12) but that doesnt mean that it will work immediately or easily
in every domain, or even that it is the best tool for a specific domain.
While some of the successes in the field have been easy, most were the

13.4 Small Changes can have Big Effects

133

result of significant effort by experienced practitioners. It is likely that for

every GP approach that has successfully solved a problem, several others
have failed. It is in the nature of academic publishing that one does not get
to hear about failures.
So, dont expect immediate success, and dont become too discouraged
by poor early results.

13.4

Small Changes can have Big Effects

Dont assume a little fiddling with parameters, operators, fitness functions, etc., is harmless. One of the awkward realities of many widely applicable tools is that they typically have numerous tunable parameters. Evolutionary algorithms such as GP are no exception. Often changing a parameter or two can have a fairly minimal impact, and averaging over many
runs is required to reliably detect those effects. Some parameter changes,
however, can produce more dramatic effects. Changing the function set, for
example, can significantly change the distribution of the sizes and shapes of
trees, especially in the early generations, and potentially bias the system in
unexpected ways.
Another source of change can be the problem domain. A common mistake is to hope that parameter settings that worked well for one problem
will also work well for what appears to be a very similar problem. Problems
that appear similar to humans, however, may have quite different search
characteristics.
In addition, there are many small differences in GP implementations that
are rarely considered important or even reported. However, our experience is
that they may produce significant changes in the behaviour of a GP system.
Differences as small as an > in place of a in an if statement can have
an important effect. For example, the substitution > may influence
the winners of tournaments, the designation of the best-of-run individual,
the choice of which elements are cloned when elitism is used, or the offspring
produced by operators which accept the offspring only if it is better or not
worse than a parent.

13.5

Big Changes can have No Effect

When big changes appear to make little difference, this can sometimes be
used to identify problems with the domain representation and fitness measure. Alternatively it may be that the problem is simply too difficult, and
no change is likely to make a significant difference.
Suppose that youre not making much progress during a set of runs. One
might react by sweeping the parameter space, doing runs with a variety of

134

13 Troubleshooting GP

different parameter settings in the hope of finding a better collection of

parameter values. What if changing the parameter values really does not
have much impact? That may mean that GP just is not able to gain any
traction given your current representation of the problem domain and fitness
function. You might, therefore, reconsider how the problem is posed to GP.
If the representation and fitness make the problem essentially a search for a
needle in a haystack, then GP will mostly be lost searching through highly
sub-optimal solutions. If so, altering parameter values is unlikely to help.
Note that essentially the same symptoms are also observed if the problem
is really beyond the capabilities of your computing resources. For example,
if the solutions are exceptionally rare, unless there are nice fitness gradients
guiding GP towards them, finding any solution will likely be beyond the
capacity of current computer technology.
How can one distinguish which is the cause of the lack of success? Is it a
bad choice of representation and fitness or is it just an extremely hard problem? To answer these questions, it is important to look at what happened
when the population size was varied. Even in the absence of fitness guidance, GP will search. In fact, it will perform a sort of random exploration
of the search space. It may not be a particularly rational exploration
we know, for example, that GP with subtree crossover tends to oversample
and re-sample short programs yet, it is still a form of stochastic search.
Thus, one may expect that, if the problem is solvable, as the population
size is progressively increased, sooner or later we should start seeing some
variation in the fitness of programs. This may be sufficient for evolution to
work with, particularly if we help it by improving the representation and
fitness function. If, instead, nothing interesting happened as the population
size was increased, then perhaps you dont have enough computing power
to solve the problem as posed, or the problem has no solution.

13.6

Study your Populations

If youre not getting your desired results, it is important to take the time
to dig around in the populations and see what is actually being evolved.2
For example, if youre using ADFs because you think that your problem
would benefit from a modular solution, examine the individuals that youre
evolving. Are they using ADFs? (Sometimes the result producing branch
simply will not refer to the ADFs at all.) Are they using them in a modular
way? Are ADFs being used multiple times? Do the ADFs encapsulate some
interesting logic, or are they just re-naming an input variable? If youre
using grammatical evolution, on the other hand, are your evolved individuals
using your grammar as you expected? Or is the grammar in fact biasing
2 If the system youre using doesnt allow you to dump individuals from a run, add
that functionality or use a different system.

13.6 Study your Populations

135

the system in an undesirable and unexpected way? Similar questions can be

asked for almost any flavour of GP; think about your goals and expectations,
and explore your populations to see to what degree those are being met.
Similarly, it can be valuable to look at the way your population changes
over time in more detail than that provided by the standard plot of fitness
vs. time. You might look at the distribution of tree sizes during your run,
or the distribution of fitness values. The distribution of fitness values might
suggest things about the structure of the search space as seen by your GP
system. If it seems to be dominated by disjoint values with large gaps
between them, then jumping those gaps may be a major challenge for your
system and it may be the cause for poor performance.
While it is important to look inside your populations, the time and effort required to do so is effectively a function of how much information is
recorded. Computer algorithms can easily generate enormous amounts of
data, especially if you produce a detailed log of events and individuals generated during your runs. Consequently, processing those results may become a
challenging data-mining exercise. Finding good ways to visualise those large
data sets can be extremely valuable. While there are a handful of papers
that specifically address visualisation, e.g., (Daida, Hilss, Ward, and Long,
2005; Pohlheim, 1999; Yamashiro, Yoshikawa, and Furuhashi, 2006), and
even the occasional workshop (Smith, Bullock, and Bird, 2002), most visualisation techniques are scattered through the literature and we are unaware
of any comprehensive review. Where we can provide a bit more guidance is
program visualisation.
An obvious (but easy to forget) advantage of GP is that we create visible
programs. This need not be the case with other approaches. So, when
presenting GP results, as a matter of routine one should consider making
a figure which contains the whole evolved program. The dot component of
the Graphviz package3 can be particularly helpful in this regard; Figure 6.2
is an example of a tree diagram generated with a simple dot input file. The
program lisp2dot4 can help with the conversion from Lisp-style expressions
to dot input files.
As the evolved trees can often be very large, it is usually helpful to perform at least some basic simplifications such as removing excess significant
digits in constants and combining constant terms. Naturally, after cleaning up the evolved program, one should make sure it still works; you should
also clearly indicate in any presentation or write-up that the program youre
presenting has been cleaned and is not the actual tree generated by GP.
There are methods to automatically simplify expressions (e.g., in Mathematica and Emacs). However, since in general there is an exponentially large
number of equivalent expressions, automatic simplification is hard. Another
3 http://www.graphviz.org/
4 http://www.cs.ucl.ac.uk/staff/W.Langdon/lisp2dot.html

136

13 Troubleshooting GP

Figure 13.1: Visualisation of the size and shape of the entire population of
1,000 individuals in the final generation of runs using a depth limit of 50 (on
the left) and a size limit of 600 (on the right). The inner circle is at depth
50, and the outer circle is at depth 100. These plots are from (Crane and
McPhee, 2005) and were drawn using the techniques described in (Daida
et al., 2005).
way is to use GP as a multi-objective evolutionary algorithm (cf. Chapter 9.)
In some cases the details of the trees are less important than their general
size and shape. Daida et al. (2005) presented a particularly useful set of
visualisation techniques for this situation.5 These techniques allow one to
see the size and shape of both individual trees as well as an aggregate view
of entire populations. Figure 13.1, for example, shows the impact of size and
depth limits on the size and shape of trees in two different runs with very
similar average sizes and depths. The plots make it clear, however, that the
shapes of the resulting trees were quite different.

13.7

Encourage Diversity

One important property to keep an eye on is population diversity. Two

particular measures that can be useful sources of information are:
Frequency of primitives Recognising when a primitive has been completely lost from the population (or its frequency has fallen to a low
level, consistent with the mutation rate) may help to diagnose problems.
5 A Mathematica implementation of this technique can be downloaded from http:
//library.wolfram.com/infocenter/MathSource/5163/.

13.8 Embrace Approximation

137

Population variety If the variety the number of distinct individuals in

the population falls below 90% of the population size, this may indicate that there is a problem. However, a high variety does not mean
the reverse. GP populations often contain introns (Section 11.3), and
so programs which are not identical may behave identically (Gustafson,
2004; McPhee and Hopper, 1999). Being different, these individuals
contribute to a high variety. So, a high variety need not indicate all
is well. Measuring phenotypic variation (i.e., diversity of behaviour)
may also be useful (McPhee, Ohs, and Hutchison, 2008).
Insufficient diversity may cause significant problems. Panmictic6 steadystate populations with tournament selection, reproduction and crossover, for
example, are prone to premature convergence. If you find this to be an issue,
measures should be taken to encourage population diversity such as:
Not using the reproduction operator.
Adding one or more mutation operators.
Using a weaker selection mechanism, e.g., using smaller tournament
sizes.
Using uniform random selection (instead of the standard negative tournaments) to decide which individuals to remove from the population.7
Using a generational population model instead of a steady-state model.
Splitting large populations into semi-isolated demes (Section 10.5).8
Using fitness sharing to encourage the formation of many fitness niches.

13.8

Embrace Approximation

There is a widespread belief that computer programs are fragile and that
any change to any bit in them will cause them to stop working. This is
fostered by the common knowledge that a small typing mistake by a human
programmer can sometimes introduce a troublesome bug into a program.
6 In a panmictic population no mating restrictions are imposed as to which individual
mates with which.
7 Doing this means that the selection scheme is no longer elitist, and it may be worthwhile to protect the best individual(s) to preserve the elitism.
8 What is meant by a large population has changed over time. In the early days
of GP, populations of 1,000 or more could be considered large. However, CPU speeds
and computer memory have increased exponentially over time. So, at the time of writing
it is not unusual to see populations of hundred of thousands or millions of individuals
being used in the solution of hard problems. Research indicates that there are benefits in
splitting populations into demes even for much smaller populations. See Section 10.5.

138

13 Troubleshooting GP

Programmers know from painful experience, however, that far from proving
immediately fatal, errors can lay hidden for years. Further, not all errors
are created equal. Some are indeed critical and must be dealt with immediately, while others are rare or largely inconsequential and so never become a
major priority. The worst are arguably the severe bugs that rarely express
themselves, as they can be extremely difficult to pin down yet still have dire
consequences when they appear.
In summary, there is no such thing as a perfect (non-trivial) humanwritten program and all such programs include a variety of errors of different
severity and with a different frequency of manifestation.9
This sort of variability is also very common in GP work. It provides the
sort of toehold that evolution can exploit in the early generations of GP
runs. The population of programs just needs to contain a few which move
vaguely in the right direction. Many of their offspring may be totally blind or
have no legs, just so long as a few continue to slime towards the light. Over
generations evolution may hopefully cobble together some useful features
from this initially unpromising ooze. The results, however, are unlikely
to be perfect or pretty. If you as a GP engineer insist on only accepting
solutions that are beautifully symmetric and walk on two legs on day one,
you are likely to be disappointed. As we have argued above, even humanwritten programs often only approximate their intended functionality. So,
why should we not accept the same from GP?
If you accept this notion, then it is important to provide your system with
some sort of gradient upon which to act, allowing it to evolve ever better
approximations. It is also important to ensure that your test environment
(usually encapsulated in the fitness function) places appropriate emphasis on
the most important features of the space from a user perspective. Consider a
problem with five test cases, four of which are fairly easy and consequently
less important, with the fifth being crucial and quite difficult. A likely
outcome in such a setting is that individuals that can do the four easier
tasks, but are unable to make the jump to the fifth. There are several
things you could try: 1) weighting the hard task more heavily, 2) dividing
it up in some way into additional sub-tasks, or 3) changing it from being a
binary condition (meaning that an individual does or does not succeed on the
fifth task) to a continuous condition, so that an individual GP program can
partially succeed on the fifth task. The first of these options is the simplest
to implement. The second two, however, create a smoother gradient for the
evolutionary process to follow, and so may yield better results.

9 This

is, of course, no excuse for writing shoddy, bug-ridden code.

13.9 Control Bloat

13.9

139

Control Bloat

If you are running out of memory or your execution times seem inordinately
high, look at how your average program size is changing over time. If programs are growing extremely fast, you may want to implement some form
of bloat control (see Section 11.3). Naturally, long runs may simply be the
result of the population being very large or the fitness evaluation being slow.
In these cases, you may find the techniques described in Chapter 10 helpful.
Controlling bloat is also important if your goal is to find a comprehensible
model, since in practice smaller models are easier to understand. A large
model will not only be difficult to understand but also may over-fit the
training data (Gelly, Teytaud, Bredeche, and Schoenauer, 2006).

13.10

Checkpoint Results

Where GP run time is long, it is important to periodically save the current

state of the run. Should the system crash, the run can be restarted from
part way through rather than at the start. Care should be taken to save the
entire state, so restarting a run does not introduce any unknown variation.
The bulk of the state to be saved is the current population.This can be
compressed, e.g., using gzip. While compression can add a few percent
to run time, reductions in disk space to less than one bit per primitive
in the population have been achieved. Checkpointing also allows you to
later continue runs that seemed particularly promising when they reached
whatever maximum generation you set initially.

13.11

Report Well

There are many potential reasons why work may be poorly received. Here
are a few: insufficient explanation of methods and algorithms, insufficient
experimental evidence, insufficient analysis, lack of statistical significance,
lack of replicability, reading too much into ones results, insufficient novelty,
poor presentation and poor English. In scientific, rather than commercial,
work it is vital to report enough details so that someone else can reproduce
your results. One very useful idea is to publish a table summarising your
GP run. Table 4.1 (page 31) contains an example tableau.
As explained in Section 13.2, it is essential to ensure that results are
statistically significant so that nobody can dismiss them as the consequence
of a lucky fluke. Complex ideas are often best explained by diagrams. When
possible, descriptions of non-trivial algorithms should be accompanied by
pseudocode, along with text describing the most important components of
the algorithm.

140

13 Troubleshooting GP

In addition to reporting your results, make sure you also discuss their
implications. If, for example, what GP has evolved means the customer can
save money or could improve their process in some way, then this should
be highlighted. Also be careful to not construct excessively complex explanations for the observations. It is very tempting to say X is probably due
to Y, but for this to be believable one should at least have made some
attempt to check if Y is indeed taking place, and whether modulations or
suppression of Y in fact produce modulations and/or suppression of X.
Finally, the most likely outcomes of a text that is badly written or badly
presented are: 1) your readers will misunderstand you, and 2) you will have
fewer readers. Spell checkers can help with typos, but whenever possible
one should ensure a native English speaker has proofread the text.

13.12

Convince your Customers

For any work in science, engineering, industry or commerce to make an impact it must be presented in a form that can convince others of the validity
of its results and conclusions. This might include: a pitch within a corporation seeking continued financial support for a project, the submission of
a research paper to a journal or the presentation of a GP-based product to
potential customers.
The burden of proof is on the users of GP, and it is important to use the
customers language. If the fact that GP discovers a particular chemical is
important in a reaction or drug design, for example, one should make this
stand out during the presentation. A great advantage of GP over many AI
techniques in that its results are often simple equations. Ensure these are
intelligible to your customer, e.g., by simplification. Also make an effort to
present your results using your customers terminology. Your GP system
may produce answers as trees, but if the customers use spreadsheets, consider translating the tree into a spreadsheet formula. Alternatively, your
customer may not be particularly interested in the details of the solution,
but instead care a great deal about which inputs the evolutionary process
tended to use.
Also, one should try to discover how the customers intend to validate
GPs answer. Do not let them invent some totally new data which has
nothing to do with the data they supplied for training (just to see how well
it does...). Avoid customers with contrived data: GP is not omnipotent
and knows nothing about things it has not seen. At the same time you
should be scrupulous about your own use of holdout data. GP is a very
powerful machine learning technique, and with this comes the ever present
danger of over-fitting. One should never allow performance on data reserved
for validation to be used to choose which answer to present to the customer.

Chapter 14

Conclusions
In his seminal paper entitled Intelligent Machinery, Turing (1948) identified three ways by which human-competitive machine intelligence might be
achieved. In connection with one of those ways, Turing said:
There is the genetical or evolutionary search by which a combination of genes is looked for, the criterion being the survival
value. (Turing, 1948)
Turing did not specify how to conduct the genetical or evolutionary
search for machine intelligence. In particular, he did not mention the idea of
a population-based parallel search in conjunction with sexual recombination
(crossover) as described in Hollands 1975 book Adaptation in Natural and
Artificial Systems (Holland, 1992, second edition). However, in Turings
paper Computing Machinery and Intelligence (Turing, 1950), he did point
out:
We cannot expect to find a good child-machine at the first attempt. One must experiment with teaching one such machine
and see how well it learns. One can then try another and see
if it is better or worse. There is an obvious connection between
this process and evolution:
Structure of the child machine = Hereditary material
Changes of the child machine = Mutations
Natural selection
= Judgement of the experimenter
In other words, Turing perceived that one possibly productive approach
to machine intelligence would involve an evolutionary process in which a
description of a computer program (the hereditary material) undergoes progressive modification (mutation) under the guidance of natural selection
(that is, selective pressure in the form of what we now call fitness).
141

142

14 Conclusions

Today, decades later, we can see that indeed Turing was right. GP has
started fulfilling his dream by providing us with a systematic method, based
on Darwinian evolution, for getting computers to automatically solve hard
real-life problems. To do so, it simply requires a high-level statement of
what needs to be done and enough computing power.
Turing also understood the need to evaluate objectively the behaviour exhibited by machines, to avoid human biases when assessing their intelligence.
This led him to propose an imitation game, now known as the Turing test for
machine intelligence, whose goals are wonderfully summarised by Samuels
position statement quoted in the introduction of this book (page 1). The
eight criteria for human competitiveness we discussed in Section 12.3 are
essentially motivated by the same goals.
At present GP is unable to produce computer programs that would pass
the full Turing test for machine intelligence, and it might not be ready
for this immense task for centuries. Nonetheless, thanks to the constant
improvements in GP technology, in its theoretical foundations and in computing power, GP has been able to solve dozens of difficult problems with
human-competitive results and to provide valuable solutions to many other
problems (see Chapter 12). These are a small step towards fulfilling Turing
and Samuels dreams, but they are also early signs of things to come. It is
reasonable to predict that in a few years time GP will be able to routinely
and competently solve important problems for us, in a variety of application
domains with human-competitive performance. Genetic programming will
then become an essential collaborator for many human activities. This will
be a remarkable step forward towards achieving true human-competitive
machine intelligence.
This field guide is an attempt to chart the terrain of techniques and
applications we have encountered in our journey in the world of genetic
programming. Much is still unmapped and undiscovered. We hope this
book will make it easier for other travellers to start many long and profitable
journeys in this exciting world.

If you have found this book to be useful, please feel free to redistribute it
(see page ii). Should you want to cite this book, please refer to the entry for
(Poli et al., 2008) in the bibliography.

Part IV

Tricks of the Trade

In the end we find that Mary does indeed have a little GP. . .
and the wolf is shown to have a very large bibliography.

143

Appendix A

Resources
The field of GP took off in the early 1990s, driven in significant part by
the publication of (Koza, 1992). Those early days were characterised by the
exponential growth common in the initial stages of successful technologies.
Many influential papers from that period can be found in the proceedings
of the International Conference on Genetic Algorithms (ICGA-93, ICGA95), the IEEE conferences on Evolutionary Computation (EC-1994), and
the Evolutionary Programming conferences. A surprisingly large number
of these are now available on-line, and weve included as many URLs as
we could in the bibliography.1 After almost twenty years, GP has matured
and is used in a wondrous array of applications from banking to betting,
from bomb detection to architectural design, from the steel industry to the
environment, from space to biology, and many others (as we have seen in
Section 12).
In 1996 it was possible to list almost all the studies and applications of
GP (Langdon, 1996), but today the range is far too great. In this appendix
we will review some of the wide variety of available sources on GP which
should assist readers who wish to explore further. Consulting information
available on the Web is certainly a good way to get quick answers for someone
who wants to know what GP is. These answers, however, will often be too
shallow for someone who really wants to then apply GP to solve practical
problems. People in this position should probably invest some time going
through more detailed accounts; some of the key books in the field include
(Banzhaf, Nordin, Keller, and Francone, 1998a; Koza, 1992; Langdon and
Poli, 2002), and others are listed in Section A.1. Technical papers in the
extensive GP literature may be the next stage. Although this literature is
easily accessible thanks to the complete on-line bibliography (Langdon et al.,
1995-2008), newcomers will often need to be selective in what they read. The
1 Each

included URL was tested and was operational at the time of writing.

145

146

A Resources

objective here may be different for different types of readers. Practitioners

may wish to focus initially on papers which deal with the same problem they
are interested in. Researchers and PhD students interested in developing a
deeper understanding of GP should also make sure they identify and read as
many seminal papers as possible, including papers or books on empirical and
theoretical studies on the inner mechanisms and behaviour of GP. These
are frequently cited in other papers and, so, can be easily identified.

A.1

Key Books

There are today more than 31 books written in English principally on GP

or its applications with more being written. These start with John Kozas
Genetic Programming 1992 (often referred to as Jaws). Koza has subsequently published three additional books on GP: Genetic Programming II:
Automatic Discovery of Reusable Programs (1994) deals with ADFs; Genetic Programming 3 (1999) covers, in particular, the evolution of analogue
circuits; Genetic Programming 4 (2003) uses GP for automatic invention.
MIT Press published three volumes in the series Advances in Genetic Programming (Angeline and Kinnear, 1996; Kinnear, 1994c; Spector, Langdon,
OReilly, and Angeline, 1999). The joint GP / genetic algorithms Kluwer
book series edited by Koza and Goldberg now contains 14 books starting
with Genetic Programming and Data Structures (Langdon, 1998). Apart
from Jaws, these tend to be for the GP specialist. The late 1990s saw the
introduction of the first textbook dedicated to GP (Banzhaf et al., 1998a).
Eiben and Smith (2003) and Goldberg (1989) provide general treatments of
evolutionary algorithms.
Other titles include: Genetic Programming (in Japanese) (Iba, 1996b),
Principia Evolvica Simulierte Evolution mit Mathematica (in German)
(Jacob, 1997) (English version (Jacob, 2001)), Data Mining Using Grammar Based Genetic Programming and Applications (Wong and Leung, 2000),
Grammatical Evolution: Evolutionary Automatic Programming in a Arbitrary Language (ONeill and Ryan, 2003), Humanoider: Sjavlarande robotar
och artificiell intelligens (in Swedish) (Nordin and Johanna, 2003), and Linear Genetic Programming (Brameier and Banzhaf, 2007).
Readers interested in mathematical and empirical analyses of GP behaviour may find Foundations of Genetic Programming (Langdon and Poli,
2002) useful.
Each of Kozas four books has an accompanying video. These videos are
now available in DVD format. Also, a small set of videos on specific GP
techniques and applications is available via on-line resources such as Google
Video and YouTube.

A.2 Key Journals

A.2

147

Key Journals

In addition to GPs own Genetic Programming and Evolvable Machines journal, Evolutionary Computation, the IEEE transaction on Evolutionary Computation, Complex Systems (Complex Systems Publication, Inc.), the new
Journal on Artificial Evolution and Applications and many others publish
GP articles. The GP bibliography (Langdon et al., 1995-2008) lists a further
375 different journals worldwide that have published articles related to GP.

A.3

Key International Meetings

EuroGP the European Conference on Genetic Programming has been

held every year since 1998. All EuroGP papers are available on line as part of
Springers LNCS series. The original annual Genetic Programming conference ran for three years (1996-1998) before combining in 1999 with the International Conference on Genetic Algorithms (ICGA) to form GECCO. 98%
of GECCO papers are available on-line. The Michigan-based Genetic Programming Theory and Practice workshop (OReilly, Yu, Riolo, and Worzel,
2004; Riolo and Worzel, 2003; Riolo, Soule, and Worzel, 2007a; Yu, Riolo,
and Worzel, 2005) has recently published its fifth proceedings (Riolo, Soule,
and Worzel, 2007b). Other EC conferences, such as CEC, PPSN, Evolution
Artificielle and WSC, also regularly contain GP papers.

A.4

GP Implementations

One of the reasons behind the success of GP is that it is easy to implement

own versions, and implementing a simple GP system from scratch remains
an excellent way to make sure one really understands the mechanics of GP.
In addition to being an exceptionally useful exercise, it is often easier to
customise (e.g., adding new, application specific genetic operators or implementing unusual, knowledge-based initialisation strategies) a system one has
built for new purposes than a large GP distribution. All of this, however,
requires reasonable programming skills and the will to thoroughly test the
resulting system until it behaves as expected.
This is actually an extremely tricky issue in highly stochastic systems
such as GP, as we discussed in Section 13.1. The problem is that almost
any system will produce interesting behaviour, but it is typically very
hard to test whether it is exhibiting the correct interesting behaviour. It
is remarkably easy for small mistakes to go unnoticed for extended periods
of time (even years).2 It is also easy to incorrectly assume that minor
2 Several years ago Nic and some of his students discovered that one of their systems
had been performing addition instead of subtraction for several months due to a copy-

148

A Resources

implementation decisions will not significantly affect the behaviour of the

system (see Section 13.4).
An alternative is to use one of the many public domain GP implementations and adapt this for ones purposes. This process is faster, and good
implementations are often robust, efficient, well documented and comprehensive. The small price to pay is the need to study the available documentation and examples. These often explain how to modify the GP system
to some extent. However, deeper modifications (such as the introduction
of new or unusual operators) will often require studying the actual source
code and a substantial amount of trial and error. Good publicly available GP implementations include: Lil-GP (Punch and Zongker, 1998), ECJ
(Luke, Panait, Balan, Paus, Skolicki, Popovici, Harrison, Bassett, Hubley,
and Chircop, 2000-2007), Open Beagle (Gagne and Parizeau, 2002) and
GPC++ (Fraser and Weinbrenner, 1993-1997). The most prominent commercial implementation remains Discipulus (RML Technologies, 1998-2007);
see (Foster, 2001) for a review. A number of older unsupported tools can be
found at ftp://cs.ucl.ac.uk/genetic/ftp.io.com/.
While the earliest GP systems were implemented in Lisp, people have
since coded GP in a huge range of different languages, including C/C++,
Java (see an example in Appendix B), JavaScript, Perl, Prolog, Mathematica, Pop-11, MATLAB, Fortran, Occam and Haskell. Typically, these evolve
expressions and programs which look like simplified Lisp. More complex target languages can be supported, however, especially with the use of more
advanced tools such as grammars and type systems (see Chapter 6). Conversely, many successful programs in machine code or low-level languages
have also climbed from the primordial ooze of initial randomness.

A.5

On-Line Resources

On-line resources appear, disappear, and move with great speed, so all the
addresses here (and elsewhere in the book), which were correct at the time
of writing, are obviously subject to change without notice after publication.
Hopefully, the most valuable resources should be readily findable using standard search tools.
One of the key on-line resources is the GP bibliography (Langdon et al.,
1995-2008).3 At the time of writing, this bibliography contains about 5,000
GP entries, roughly half of which can be downloaded immediately.4
paste error. Fortunately no published results were affected, but it was a very unsettling
experience.
3 http://www.cs.bham.ac.uk/~wbl/biblio/
4 The GP bibliography is a volunteer effort and depends crucially on submissions from
users. Authors are encouraged to check that their GP publications are listed, and send
missing entries to the bibliographys maintainers.

A.5 On-Line Resources

149

The GP bibliography has a variety of interfaces, including a graphical

representation of GPs collaborative network (see Figure A.1). The bibliography allows for quick jumps between papers linked by authors and allows
one to sort the author list by the number of GP publications. Full references are provided in both BibTEX and Refer formats for direct inclusion
in papers written in LATEX and Microsoft Word, respectively. The GP bibliography is also part of the Collection of Computer Sciences Bibliographies
(Achilles and Ortyl, 1995-2008), which provides a comprehensive Lucerne
syntax search engine.
From early on there has been an active, open email discussion list: the
Genetic Programming mailing list (2001-2008). The EC-Digest (1985-2008)
is a moderated list covering evolutionary computation more broadly, and
often contains GP related announcements.
Kozas http://www.genetic-programming.org/ contains a ton of useful information for the novice, including a short tutorial on What is Genetic
Programming and the Lisp implementation of GP from Genetic Programming (Koza, 1992).

150

A Resources

Figure A.1: Co-authorship connections within GP. Each of the 1,141 dots
indicates an author, and edges link people who have co-authored one or
more GP papers. (To reduce clutter only links to first authors are shown.)
The size of each dot indicates the number of entries. The on-line version is
annotated using JavaScript and contains hyperlinks to authors and their
GP papers. The graph was created by GraphViz twopi, which tries to
place strongly connected people close together. This diagram displays just
the centrally connected component (Tomassini et al., 2007) and contains
approximately half of all GP papers. The remaining papers are not linked
by co-authorship to this graph. Several other large components are also
available on-line via the GP Bibliography (Langdon et al., 1995-2008).

Appendix B

TinyGP
TinyGP1 i s a highly optimised GP system that was originally developed to
meet the specifications set out in the TinyGP competition of the Genetic and
Evolutionary Computation Conference (GECCO) 2004. We include it as a
working example of a real GP system, to show that GP software tools are
not necessarily big, complex and difficult to understand. The system can be
used as is or can be modified or extended for a users specific applications.
Furthermore, TinyGP may serve as a guide to other implementations of
genetic programming.
The following section provides a description of the main characteristics
of TinyGP. Section B.2 describes the format for the input files for TinyGP.
Section B.3 provides further details on the implementation and the source
code for a Java version of TinyGP. Finally, Section B.4 describes a sample
run of the system.
There are numerous other GP systems available on-line. See Section A.4
for a discussion of some of the options.

B.1

Overview of TinyGP

TinyGP is a symbolic regression system with the following characteristics:

1. The terminal set includes a user-definable number of floating point
variables (named X1 to XN).
2. The function set includes multiplication, protected division, subtraction and addition.
3. The fitness cases are read from a file (the format is given below).
1 http://cswww.essex.ac.uk/staff/rpoli/TinyGP/

151

152

B TinyGP

4. The system is steady state. A generation is considered concluded

when POPSIZE (see below) crossover/mutation events have been performed.
5. Selection is performed using tournament selection.
6. Negative tournaments are used for the selection of the individuals to
be replaced at each steady-state-GP iteration.
7. Subtree crossover is used. The selection of crossover points is uniform,
so every node is chosen equally likely.
8. Point mutation is used. That is, points (nodes) in the tree are randomly chosen. If a point is a terminal, then it is replaced by another
randomly chosen terminal. If it is a function, then it is replaced by
another randomly chosen function with the same number of inputs.
9. The following parameters are implemented as static class variables:
The maximum length any GP program can take: MAX LEN.
The size of the population: POPSIZE.
The maximum depth initial programs can have: DEPTH. Note 0
represents the depth of programs containing just one terminal.
The maximum number of generations allowed for a run:
GENERATIONS.
The
probability
of
creating
new
individuals
via
crossover:
CROSSOVER PROB. The mutation probability is
1 - CROSSOVER PROB.
The mutation probability (per node) when point mutation is chosen as the variation operator: PMUT PER NODE.
The tournament size: TSIZE.
10. The parameters and the random seed are printed when each run starts.
11. The fitness function is minus the sum of the absolute differences between the actual program output and the desired output for each fitness case. TinyGP maximises it.
12. The grow initialisation method is used to create the initial population.
13. At each generation the following statistics are calculated and printed:
The generation number.
The average fitness of the individuals in the population.
The fitness of the best individual in the population.

B.2 Input Data Files for TinyGP

153

The average size of the programs in the current generation.

The best individual in the population.
14. The random number generator can be seeded via the command line.
If this command line parameter is absent, the system uses the current
time to seed the random number generator.
15. The name of the file containing the fitness cases can be passed to
the system via the command line. If the command line parameter is
absent, the system assumes the data are stored in the current directory
in a file called problem.dat.
16. If the total error made by the best program goes below 105 TinyGP
prints a message indicating success and stops. If the problem has
not been solved after the maximum number of generations, it prints a
message indicating failure and stops.

B.2

Input Data Files for TinyGP

The input files for TinyGP have the following plain ASCII format:
HEADER
// See below
FITNESSCASE1 // The f i t n e s s c a s e s ( one p e r l i n e )
FITNESSCASE2
FITNESSCASE3
....
Each fitness case is of the form
X1 . . . XN TARGET
where X1 to XN represent a set of input values for a program, while
TARGET represents the desired output for the given inputs.
The header has the following entries
NVAR NRAND MINRAND MAXRAND NFITCASES
where NVAR is an integer representing the number of variables the system
should use, NRAND is an integer representing the number of random constants to be provided in the primitive set, MINRAND is a float representing
the lower limit of the range used to generate random constants, MAXRAND
is the corresponding upper limit, and NFITCASES is an integer representing the number of fitness cases. NRAND can be set to 0, in which case
MINRAND and MAXRAND are ignored. For example:
1 100 -5 5 63
0.0 0

154

B TinyGP

0.1 0.0998334166468282
0.2 0.198669330795061
0.3 0.29552020666134
....
55 LINES OMITTED
....
5.9 -0.373876664830236
6.0 -0.279415498198926
6.1 -0.182162504272095
6.2 -0.0830894028174964
These fitness cases are sin(x) for x {0.0, 0.1, 0.2, . . . 6.2}

B.3

Source Code

The original TinyGP system was implemented, in the C programming language, to maximise efficiency and minimise the size of the executable.2 The
version presented here is a Java re-implementation of TinyGP. The original
version did not allow the use of random numerical constants.
How does TinyGP work? The system is based on the standard flattened
(linear) representation for trees, which effectively corresponds to listing the
primitives in prefix notation but without any brackets. Each primitive occupies one byte. A program is simply a vector of characters. The parameters
of the system are as specified in Section B.1. They are fixed at compile time
through a series of static class variable assignments. The operators used
are subtree crossover and point mutation. The selection of the crossover
points is performed at random with uniform probability. The primitive set
and fitness function are as indicated above. The code uses recursion for the
creation of the initial population (grow), for the identification of the subtree
rooted at a particular crossover point (traverse), for program execution
(run), and for printing programs (print indiv). A small number of global
variables have been used. For example, the variable program is a program
counter used during the recursive interpretation of programs, which is automatically incremented every time a primitive is evaluated. Although using
global variables is normally considered bad programming practice, this was
done purposely, after extensive experimentation, to reduce the executables
size.
2 The C version of TinyGP is probably the worlds smallest tree-based symbolicregression GP system. The source code, in C, is 5,906 bytes. The original version included
a compilation script which, with a variety of tricks, created a self-extracting executable
occupying 2,871 bytes (while the actual size of the executable after self-extraction was
4,540 bytes). All optimisations in the code were aimed at bringing the executable size
(as opposed to the source code size) down, the main purpose being to show that, against
popular belief, it is possible to have really tiny and efficient GP systems.

B.3 Source Code

155

The code reads command line arguments using the standard args array.
Generally the code is quite standard and should be self-explanatory for
anyone who can program in Java, whether or not they have implemented a
GP system before. Therefore very few comments have been provided in the
source code.
The source is provided below.
1
2
3
4
5
6

/
Program:

Author :

tiny gp . java
Riccardo Poli (email : rpoli@essex . ac .uk)

7
8
9
10

import j a v a . u t i l . ;
import j a v a . i o . ;
import j a v a . t e x t . DecimalFormat ;

11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40

public c l a s s t i n y g p {
double [ ] f i t n e s s ;
char [ ] [ ] pop ;
s t a t i c Random rd = new Random ( ) ;
static f i n a l int
ADD = 1 1 0 ,
SUB = 1 1 1 ,
MUL = 1 1 2 ,
DIV = 1 1 3 ,
FSET START = ADD,
FSET END = DIV ;
s t a t i c double [ ] x = new double [ FSET START ] ;
s t a t i c double minrandom , maxrandom ;
s t a t i c char [ ] program ;
s t a t i c i n t PC;
s t a t i c i n t varnumber , f i t n e s s c a s e s , randomnumber ;
s t a t i c double f b e s t p o p = 0 . 0 , f av g p op = 0 . 0 ;
s t a t i c long s e e d ;
s t a t i c double a v g l e n ;
static f i n a l int
MAX LEN = 1 0 0 0 0 ,
POPSIZE = 1 0 0 0 0 0 ,
DEPTH
= 5,
GENERATIONS = 1 0 0 ,
TSIZE = 2 ;
public s t a t i c f i n a l double
PMUT PER NODE = 0 . 0 5 ,
CROSSOVER PROB = 0 . 9 ;
s t a t i c double [ ] [ ] t a r g e t s ;

41
42
43
44
45
46

double run ( ) { / Interpreter /

char p r i m i t i v e = program [ PC++];
i f ( p r i m i t i v e < FSET START )
return ( x [ p r i m i t i v e ] ) ;
switch ( p r i m i t i v e ) {

156
case ADD : return ( run ( ) + run ( ) ) ;
case SUB : return ( run ( ) run ( ) ) ;
case MUL : return ( run ( ) run ( ) ) ;
case DIV : {
double num = run ( ) , den = run ( ) ;
i f ( Math . abs ( den ) <= 0 . 0 0 1 )
return ( num ) ;
else
return ( num / den ) ;
}
}
return ( 0 . 0 ) ; // should never get here

47
48
49
50
51
52
53
54
55
56
57
58
59

B TinyGP

60
61
62
63

i n t t r a v e r s e ( char [ ] b u f f e r , i n t b u f f e r c o u n t ) {
i f ( b u f f e r [ b u f f e r c o u n t ] < FSET START )
return ( ++b u f f e r c o u n t ) ;

64

switch ( b u f f e r [ b u f f e r c o u n t ] ) {
case ADD:
case SUB :
case MUL:
case DIV :
return ( t r a v e r s e ( b u f f e r , t r a v e r s e ( b u f f e r , ++b u f f e r c o u n t
) ) );
}
return ( 0 ) ; // should never get here

65
66
67
68
69
70

71
72
73

74
75
76
77
78

void s e t u p f i t n e s s ( S t r i n g fname ) {
try {
int i , j ;
String line ;

79
80
81
82
83
84
85
86
87

88

89

90

91
92
93

BufferedReader in =
new B u f f e r e d R e a d e r (
new
F i l e R e a d e r ( fname ) ) ;
l i n e = in . readLine () ;
S t r i n g T o k e n i z e r t o k e n s = new S t r i n g T o k e n i z e r ( l i n e ) ;
varnumber = I n t e g e r . p a r s e I n t ( t o k e n s . nextToken ( ) . t r i m ( ) ) ;
randomnumber = I n t e g e r . p a r s e I n t ( t o k e n s . nextToken ( ) . t r i m ( ) )
;
minrandom =
Double . p a r s e D o u b l e ( t o k e n s . nextToken ( ) .
trim ( ) ) ;
maxrandom = Double . p a r s e D o u b l e ( t o k e n s . nextToken ( ) . t r i m ( ) )
;
f i t n e s s c a s e s = I n t e g e r . p a r s e I n t ( t o k e n s . nextToken ( ) . t r i m ( ) )
;
t a r g e t s = new double [ f i t n e s s c a s e s ] [ varnumber + 1 ] ;
i f ( varnumber + randomnumber >= FSET START )
System . out . p r i n t l n ( t o o many v a r i a b l e s and c o n s t a n t s ) ;

94
95

f o r ( i = 0 ; i < f i t n e s s c a s e s ; i ++ ) {

B.3 Source Code

l i n e = in . readLine () ;
t o k e n s = new S t r i n g T o k e n i z e r ( l i n e ) ;
f o r ( j = 0 ; j <= varnumber ; j ++) {
t a r g e t s [ i ] [ j ] = Double . p a r s e D o u b l e ( t o k e n s . nextToken ( ) .
trim ( ) ) ;
}

96
97
98
99

100

}
in . close () ;

101
102

}
catch ( F i l e N o t F o u n d E x c e p t i o n e ) {
System . out . p r i n t l n ( ERROR: P l e a s e p r o v i d e a data f i l e ) ;
System . e x i t ( 0 ) ;
}
catch ( E x c e p t i o n e ) {
System . out . p r i n t l n ( ERROR: I n c o r r e c t data f o r m a t ) ;
System . e x i t ( 0 ) ;
}

103
104
105
106
107
108
109
110
111
112

113
114
115
116

double f i t n e s s f u n c t i o n ( char [ ] Prog ) {

int i = 0 , l e n ;
double r e s u l t , f i t = 0 . 0 ;

117

l e n = t r a v e r s e ( Prog , 0 ) ;
f o r ( i = 0 ; i < f i t n e s s c a s e s ; i ++ ) {
f o r ( i n t j = 0 ; j < varnumber ; j ++ )
x [ j ] = targets [ i ] [ j ] ;
program = Prog ;
PC = 0 ;
r e s u l t = run ( ) ;
f i t += Math . abs ( r e s u l t t a r g e t s [ i ] [ varnumber ] ) ;
}
return( f i t ) ;

118
119
120
121
122
123
124
125
126
127
128

129
130
131

i n t grow ( char [ ] b u f f e r , i n t pos , i n t max , i n t depth ) {

char prim = ( char ) rd . n e x t I n t ( 2 ) ;

132
133
134

i f ( pos >= max )

return ( 1 ) ;

135
136
137

i f ( pos == 0 )
prim = 1 ;

138
139
140
141
142
143
144
145

146
147
148

157

i f ( prim == 0 | | depth == 0 ) {
prim = ( char ) rd . n e x t I n t ( varnumber + randomnumber ) ;
b u f f e r [ pos ] = prim ;
return ( pos +1) ;
}
else {
prim = ( char ) ( rd . n e x t I n t (FSET END FSET START + 1 ) +
FSET START) ;
switch ( prim ) {
case ADD:
case SUB :

158

case MUL:
case DIV :
b u f f e r [ pos ] = prim ;
return ( grow ( b u f f e r , grow ( b u f f e r , pos +1 , max , depth 1) ,
max , depth 1 ) ) ;
}

149
150
151
152
153
154

}
return ( 0 ) ;

155
156
157

B TinyGP

// should never get here

158
159
160
161
162
163

164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189

i n t p r i n t i n d i v ( char [ ] b u f f e r , i n t b u f f e r c o u n t e r ) {
i n t a1 =0 , a2 ;
i f ( b u f f e r [ b u f f e r c o u n t e r ] < FSET START ) {
i f ( b u f f e r [ b u f f e r c o u n t e r ] < varnumber )
System . out . p r i n t ( X+ ( b u f f e r [ b u f f e r c o u n t e r ] + 1 )+
);
else
System . out . p r i n t ( x [ b u f f e r [ b u f f e r c o u n t e r ] ] ) ;
return ( ++b u f f e r c o u n t e r ) ;
}
switch ( b u f f e r [ b u f f e r c o u n t e r ] ) {
case ADD: System . out . p r i n t ( ( ) ;
a1=p r i n t i n d i v ( b u f f e r , ++b u f f e r c o u n t e r ) ;
System . out . p r i n t ( + ) ;
break ;
case SUB : System . out . p r i n t ( ( ) ;
a1=p r i n t i n d i v ( b u f f e r , ++b u f f e r c o u n t e r ) ;
System . out . p r i n t ( ) ;
break ;
case MUL: System . out . p r i n t ( ( ) ;
a1=p r i n t i n d i v ( b u f f e r , ++b u f f e r c o u n t e r ) ;
System . out . p r i n t ( ) ;
break ;
case DIV : System . out . p r i n t ( ( ) ;
a1=p r i n t i n d i v ( b u f f e r , ++b u f f e r c o u n t e r ) ;
System . out . p r i n t ( / ) ;
break ;
}
a2=p r i n t i n d i v ( b u f f e r , a1 ) ;
System . out . p r i n t ( ) ) ;
return ( a2 ) ;
}

190
191
192
193
194
195

s t a t i c char [ ] b u f f e r = new char [MAX LEN ] ;

char [ ] c r e a t e r a n d o m i n d i v ( i n t depth ) {
char [ ] i n d ;
int l e n ;

196
197

l e n = grow ( b u f f e r , 0 , MAX LEN, depth ) ;

198
199
200
201

while ( l e n < 0 )
l e n = grow ( b u f f e r , 0 , MAX LEN, depth ) ;

B.3 Source Code

159

i n d = new char [ l e n ] ;

202
203

System . a r r a y c o p y ( b u f f e r , 0 , ind , 0 , l e n ) ;
return ( i n d ) ;

204
205
206

207
208

209
210

char [ ] [ ] c r e a t e r a n d o m p o p ( i n t n , i n t depth , double [ ]

fitness ) {
char [ ] [ ] pop = new char [ n ] [ ] ;
int i ;

211

f o r ( i = 0 ; i < n ; i ++ ) {
pop [ i ] = c r e a t e r a n d o m i n d i v ( depth ) ;
f i t n e s s [ i ] = f i t n e s s f u n c t i o n ( pop [ i ] ) ;
}
return ( pop ) ;

212
213
214
215
216
217

218
219
220
221
222
223
224

void s t a t s ( double [ ] f i t n e s s , char [ ] [ ] pop , i n t gen ) {

i n t i , b e s t = rd . n e x t I n t (POPSIZE) ;
int node count = 0 ;
fbestpop = f i t n e s s [ best ] ;
fa vgp o p = 0 . 0 ;

225

f o r ( i = 0 ; i < POPSIZE ; i ++ ) {
n o d e c o u n t += t r a v e r s e ( pop [ i ] , 0 ) ;
fa v gp o p += f i t n e s s [ i ] ;
i f ( f i t n e s s [ i ] > fbestpop ) {
best = i ;
fbestpop = f i t n e s s [ i ] ;
}
}
a v g l e n = ( double ) n o d e c o u n t / POPSIZE ;
fa vgp o p /= POPSIZE ;
System . out . p r i n t ( G e n e r a t i o n=+gen+ Avg F i t n e s s=+(fa v gp o p
)+
Best F i t n e s s=+(f b e s t p o p )+ Avg S i z e=+
a v g l e n+
\ nBest I n d i v i d u a l : ) ;
p r i n t i n d i v ( pop [ b e s t ] , 0 ) ;
System . out . p r i n t ( \n ) ;
System . out . f l u s h ( ) ;

226
227
228
229
230
231
232
233
234
235
236

237

238
239
240
241
242

243
244
245
246

i n t tournament ( double [ ] f i t n e s s , i n t t s i z e ) {
i n t b e s t = rd . n e x t I n t (POPSIZE) , i , c o m p e t i t o r ;
double f b e s t = 1.0 e34 ;

247
248
249
250
251
252
253

f o r ( i = 0 ; i < t s i z e ; i ++ ) {
c o m p e t i t o r = rd . n e x t I n t (POPSIZE) ;
i f ( f i t n e s s [ competitor ] > f b e s t ) {
f b e s t = f i t n e s s [ competitor ] ;
best = competitor ;
}

160
}
return ( b e s t ) ;

254
255
256

B TinyGP

257
258
259
260

i n t n e g a t i v e t o u r n a m e n t ( double [ ] f i t n e s s , i n t t s i z e ) {
i n t w o r s t = rd . n e x t I n t (POPSIZE) , i , c o m p e t i t o r ;
double f w o r s t = 1 e34 ;

261

f o r ( i = 0 ; i < t s i z e ; i ++ ) {
c o m p e t i t o r = rd . n e x t I n t (POPSIZE) ;
i f ( f i t n e s s [ competitor ] < fworst ) {
fworst = f i t n e s s [ competitor ] ;
worst = competitor ;
}
}
return ( w o r s t ) ;

262
263
264
265
266
267
268
269
270

271
272
273
274
275
276
277

char [ ] c r o s s o v e r ( char [ ] p a r e n t 1 , char [ ] p a r e n t 2 ) {

i n t x o 1 s t a r t , xo1end , x o 2 s t a r t , xo2end ;
char [ ] o f f s p r i n g ;
int le n1 = t r a v e r s e ( parent1 , 0 ) ;
int le n2 = t r a v e r s e ( parent2 , 0 ) ;
int l e n o f f ;

278

x o 1 s t a r t = rd . n e x t I n t ( l e n 1 ) ;
xo1end = t r a v e r s e ( p a r e n t 1 , x o 1 s t a r t ) ;

279
280
281

x o 2 s t a r t = rd . n e x t I n t ( l e n 2 ) ;
xo2end = t r a v e r s e ( p a r e n t 2 , x o 2 s t a r t ) ;

282
283
284

l e n o f f = x o 1 s t a r t + ( xo2end x o 2 s t a r t ) + ( l e n 1 xo1end ) ;

285
286

o f f s p r i n g = new char [ l e n o f f ] ;

287
288

System . a r r a y c o p y ( p a r e n t 1 , 0 , o f f s p r i n g , 0 , x o 1 s t a r t ) ;
System . a r r a y c o p y ( p a r e n t 2 , x o 2 s t a r t , o f f s p r i n g , x o 1 s t a r t ,
( xo2end x o 2 s t a r t ) ) ;
System . a r r a y c o p y ( p a r e n t 1 , xo1end , o f f s p r i n g ,
x o 1 s t a r t + ( xo2end x o 2 s t a r t ) ,
( l e n 1 xo1end ) ) ;

289
290
291
292
293
294
295

return ( o f f s p r i n g ) ;

296
297

298
299
300
301
302

char [ ] mutation ( char [ ] p a r e n t , double pmut ) {

int l e n = t r a v e r s e ( parent , 0 ) , i ;
int mutsite ;
char [ ] p a r e n t c o p y = new char [ l e n ] ;

303
304
305
306
307
308

System . a r r a y c o p y ( p a r e n t , 0 , p a r e n t c o p y , 0 , l e n ) ;
f o r ( i = 0 ; i < l e n ; i ++ ) {
i f ( rd . nextDouble ( ) < pmut ) {
mutsite = i ;
i f ( p a r e n t c o p y [ m u t s i t e ] < FSET START )

B.3 Source Code

p a r e n t c o p y [ m u t s i t e ] = ( char ) rd . n e x t I n t ( varnumber ) ;
else
switch ( p a r e n t c o p y [ m u t s i t e ] ) {
case ADD:
case SUB :
case MUL:
case DIV :
parentcopy [ mutsite ] =
( char ) ( rd . n e x t I n t (FSET END FSET START + 1 )
+ FSET START) ;
}
}

309
310
311
312
313
314
315
316
317
318
319
320

}
return ( p a r e n t c o p y ) ;

321
322
323

324
325
326
327
328
329
330
331
332
333
334
335
336

void p r i n t p a r m s ( ) {
System . out . p r i n t ( TINY GP ( Java v e r s i o n ) \n ) ;
System . out . p r i n t ( SEED=+s e e d+ \nMAX LEN=+MAX LEN+
\nPOPSIZE=+POPSIZE+ \nDEPTH=+DEPTH+
\nCROSSOVER PROB=+CROSSOVER PROB+
\nPMUT PER NODE=+PMUT PER NODE+
\nMIN RANDOM=+minrandom+
\nMAX RANDOM=+maxrandom+
\nGENERATIONS=+GENERATIONS+
\nTSIZE=+TSIZE+
\n\n ) ;
}

337
338
339
340
341
342
343
344
345
346
347

public t i n y g p ( S t r i n g fname , long s ) {

f i t n e s s = new double [ POPSIZE ] ;
seed = s ;
i f ( s e e d >= 0 )
rd . s e t S e e d ( s e e d ) ;
s e t u p f i t n e s s ( fname ) ;
pop = c r e a t e r a n d o m p o p (POPSIZE , DEPTH, f i t n e s s ) ;
f o r ( i n t i = 0 ; i < FSET START ; i ++ )
x [ i ]= ( maxrandomminrandom ) rd . nextDouble ( )+minrandom ;
}

348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363

void e v o l v e ( ) {
i n t gen = 0 , i n d i v s , o f f s p r i n g , p a r e n t 1 , p a r e n t 2 , p a r e n t ;
double n e w f i t ;
char [ ] newind ;
print parms () ;
s t a t s ( f i t n e s s , pop , 0 ) ;
f o r ( gen = 1 ; gen < GENERATIONS; gen ++ ) {
if (
f b e s t p o p > 1e5 ) {
System . out . p r i n t ( PROBLEM SOLVED\n ) ;
System . e x i t ( 0 ) ;
}
f o r ( i n d i v s = 0 ; i n d i v s < POPSIZE ; i n d i v s ++ ) {
i f ( rd . nextDouble ( ) > CROSSOVER PROB ) {
p a r e n t 1 = tournament ( f i t n e s s , TSIZE ) ;
p a r e n t 2 = tournament ( f i t n e s s , TSIZE ) ;

161

162

B TinyGP
newind = c r o s s o v e r ( pop [ p a r e n t 1 ] , pop [ p a r e n t 2 ] ) ;
}
else {
p a r e n t = tournament ( f i t n e s s , TSIZE ) ;
newind = mutation ( pop [ p a r e n t ] , PMUT PER NODE ) ;
}
n e w f i t = f i t n e s s f u n c t i o n ( newind ) ;
o f f s p r i n g = n e g a t i v e t o u r n a m e n t ( f i t n e s s , TSIZE ) ;
pop [ o f f s p r i n g ] = newind ;
f i t n e s s [ offspring ] = newfit ;
}
s t a t s ( f i t n e s s , pop , gen ) ;

364
365
366
367
368
369
370
371
372
373
374
375

}
System . out . p r i n t ( PROBLEM NOT SOLVED\n ) ;
System . e x i t ( 1 ) ;

376
377
378

379
380

public s t a t i c void main ( S t r i n g [ ] a r g s ) {

S t r i n g fname = problem . dat ;
long s = 1;

381
382
383
384

i f ( a r g s . l e n g t h == 2 ) {
s = I n t e g e r . valueOf ( args [ 0 ] ) . intValue ( ) ;
fname = a r g s [ 1 ] ;
}
i f ( a r g s . l e n g t h == 1 ) {
fname = a r g s [ 0 ] ;
}

385
386
387
388
389
390
391
392

t i n y g p gp = new t i n y g p ( fname , s ) ;
gp . e v o l v e ( ) ;

393
394

395
396

};

B.4

Compiling and Running TinyGP

It is very common, nowadays, for people to write and execute code within
some development environment. Each has its own way of doing these operations, but the process is typically very straightforward.
If one wants to compile TinyGP from the operating systems shell, this
can be done by issuing the command javac -O tiny gp.java. This applies
to both Unix and Windows users. Windows users will have to click on
StartRun and then issue the command cmd to launch a shell. Of course,
if the javac Java compiler and/or the tiny gp.java source file are not in
the current directory/folder, then full path names must be provided when
issuing the compilation command.
If the dataset is stored in a file problem.dat, the program can then
simply be launched with the command java tiny gp. Otherwise, the user
can specify a different datafile on the command line, by giving the command

B.4 Compiling and Running TinyGP

163

java tiny gp FILE, where FILE is the dataset file name (which can include
the full path to the file). Finally, the user can specify both the datafile and
a seed for the random number generator on the command line, by giving
the command java tiny gp SEED FILE, where SEED is an integer.
As an example, we ran TinyGP on the sin(x) dataset described in Section B.2 (which is available at http://cswww.essex.ac.uk/staff/rpoli/
TinyGP/sin-data.txt). The output produced by the program was something like the following
TINY GP ( Java v e r s i o n )
SEED=1
MAX LEN=10000
POPSIZE=100000
DEPTH=5
CROSSOVER PROB=0.9
PMUT PER NODE=0.05
MIN RANDOM=5.0
MAXRANDOM=5.0
GENERATIONS=100
TSIZE=2

G e n e r a t i o n=0 Avg F i t n e s s =42.53760218120066 Best F i t n e s s

=39.997953686554816 Avg S i z e =10.9804
Best I n d i v i d u a l : ( 1 . 5 8 9 8 1 6 3 3 4 4 5 8 0 5 5 / 2.128280559500907)
G e n e r a t i o n=1 Avg F i t n e s s =1226.404415960088 Best F i t n e s s
=24.441994244449372 Avg S i z e =10.97024
Best I n d i v i d u a l : ( ( ( 0 . 3 8 3 9 8 6 7 9 4 4 2 2 2 2 0 6 / 2 .27 9612716 24 28 4 03 ) +
( 1.83868 12853617673 / 1.06553859601892) )
(((4.984026635222818 (0.17196413319878445
0 . 1 2 9 4 0 4 4 2 1 5 6 5 5 9 2 3 ) ) + (X1 1.8956001614031734) )
0.3627020733460027) )
...

The flip-o-rama animation in the footer of the bibliography and index

include plots of the best and mean fitness, the mean program size and the
behaviour of the best-so-far individual at each generation. The animation
should be viewed by rapidly flipping the pages of the book from the beginning of the bibliography onward. For convenience, the plots corresponding
to the final generation are also reported in Figure B.1.
As one can see, GP progressively evolves better and better approximations to the sine function. The best individual at the end of the run had an
error of 1.88. Its unsimplified version as produced by the system is
(X1

/ ( ( 2 . 7 6 6 0 9 7 8 9 9 9 5 4 3 8 3 (X1 / ( ( ( X1 / ( ( ( ( X1 / (X1
3.200 1163 763204445) ) X1 ) 3.2001163763204445 )
3.200 1163 763204445) ) + X1 ) + (X1 (X1
3 . 9 5 3 2 4 3 6 9 3 8 9 5 4 3 7 6 ) ) ) ) ) ( ( ( X1 X1 ) / ( ( ( X1 /
( 3 . 9 5 3 2 4 3 6 9 3 8 9 5 4 3 7 6 3.20011637632 04 445) ) X1 )
3.200 1163 763204445) ) / ( ( ( X1 + X1 ) / (X1 X1 ) ) + X1 ) )
))

164

B TinyGP
1000

Avg Fitness
Best Fitness

Fitness

100

10

20

40

60

80

100

Generations
100

Avg Size

90
80

Average Size

70
60
50
40
30
20
10

20

40

60

80

100

Generations
2

sin(x)
GP (gen=99)

1.5

0.5

-0.5

-1

-1.5

-2

Figure B.1: Final generation of a TinyGP sample run: best and mean
fitness (top), mean program size (middle) and behaviour of the best-so-far
individual (bottom).

B.4 Compiling and Running TinyGP

165

which can be simplified to

ax
x
xb +x+x(xc)

x2

2
d xe

( x2 +x)

where
a = 2.76609789995
b = 10.240744822
c = 3.9532436939
d = 3.20011637632
e = 12.6508398844
Hardly an obvious approximation for the sine function, but still a very accurate one, at least over the test range.

Bibliography
H. Abbass, N. Hoai, and R. McKay. Anttag: A new method to compose computer
programs using colonies of ants. In IEEE Congress on Evolutionary Computation,
2002., 2002. URL http://citeseer.ist.psu.edu/abbass02anttag.html.
A.-C. Achilles and P. Ortyl. The Collection of Computer Science Bibliographies, 19952008. URL http://liinwww.ira.uka.de/bibliography/.
G. Adorni, S. Cagnoni, and M. Mordonini. Efficient low-level vision program design using sub-machine-code genetic programming.
In M. Gori, editor, AIIA
2002, Workshop sulla Percezione e Visione nelle Macchine, Siena, Italy, 10-13
September 2002. URL http://www-dii.ing.unisi.it/aiia2002/paper/PERCEVISIO/
adorni-aiia02.pdf.
GPBiB
A. Agapitos, J. Togelius, and S. M. Lucas. Multiobjective techniques for the use of state
in genetic programming applied to simulated car racing. In D. Srinivasan and L. Wang,
editors, 2007 IEEE Congress on Evolutionary Computation, pages 15621569, Singapore, 25-28 September 2007. IEEE Computational Intelligence Society, IEEE Press.
ISBN 1-4244-1340-0.
GPBiB
S. H. Al-Sakran, J. R. Koza, and L. W. Jones. Automated re-invention of a previously
patented optical lens system using genetic programming. In M. Keijzer, et al., editors,
Proceedings of the 8th European Conference on Genetic Programming, volume 3447 of
Lecture Notes in Computer Science, pages 2537, Lausanne, Switzerland, 30 March 1 April 2005. Springer. ISBN 3-540-25436-6. URL http://springerlink.metapress.
com/openurl.asp?genre=article&issn=0302-9743&volume=3447&spage=25.
GPBiB
R. Aler, D. Borrajo, and P. Isasi. Using genetic programming to learn and improve
control knowledge. Artificial Intelligence, 141(1-2):2956, October 2002. URL http:
//scalab.uc3m.es/~dborrajo/papers/aij-evock.ps.gz.
GPBiB
J. Allen, H. M. Davey, D. Broadhurst, J. K. Heald, J. J. Rowland, S. G. Oliver, and
D. B. Kell. High-throughput classification of yeast mutants for functional genomics
using metabolic footprinting. Nature Biotechnology, 21(6):692696, June 2003. URL
http://dbkgroup.org/Papers/NatureBiotechnology21(692-696).pdf.
GPBiB
L. Alonso and R. Schott. Random Generation of Trees. Kluwer Academic Publishers,
Boston, MA, USA, 1995. ISBN 0-7923-9528-X.
L. Altenberg. Emergent phenomena in genetic programming. In A. V. Sebald and L. J.
Fogel, editors, Evolutionary Programming Proceedings of the Third Annual Conference, pages 233241, San Diego, CA, USA, 24-26 February 1994. World Scientific Publishing. ISBN 981-02-1810-9. URL http://dynamics.org/~altenber/PAPERS/EPIGP/.
GPBiB

167

168

BIBLIOGRAPHY

A. P. Alves da Silva and P. J. Abrao. Applications of evolutionary computation in electric

power systems. In D. B. Fogel, et al., editors, Proceedings of the 2002 Congress on
Evolutionary Computation CEC2002, pages 10571062. IEEE Press, 2002. ISBN 07803-7278-6.
GPBiB
D. Ando, P. Dahlsted, M. Nordahl, and H. Iba. Interactive GP with tree representation
of classical music pieces. In M. Giacobini, et al., editors, Applications of Evolutionary
Computing, EvoWorkshops2007: EvoCOMNET, EvoFIN, EvoIASP, EvoInteraction,
EvoMUSART, EvoSTOC, EvoTransLog, volume 4448 of LNCS, pages 577584, Valencia, Spain, 11-13 April 2007. Springer Verlag.
GPBiB
D. Andre, F. H. Bennett, III, and J. R. Koza. Discovery by genetic programming of a
cellular automata rule that is better than any known rule for the majority classification
problem. In J. R. Koza, et al., editors, Genetic Programming 1996: Proceedings of
the First Annual Conference, pages 311, Stanford University, CA, USA, 2831 July
1996. MIT Press. URL http://www.genetic-programming.com/jkpdf/gp1996gkl.pdf.
GPBiB
D. Andre and J. R. Koza. Parallel genetic programming: A scalable implementation
using the transputer network architecture. In P. J. Angeline and K. E. Kinnear, Jr.,
editors, Advances in Genetic Programming 2, chapter 16, pages 317338. MIT Press,
Cambridge, MA, USA, 1996. ISBN 0-262-01158-1.
GPBiB
D. Andre and J. R. Koza. A parallel implementation of genetic programming that achieves
super-linear performance. Information Sciences, 106(3-4):201218, 1998. ISSN 00200255. URL http://www.sciencedirect.com/science/article/B6V0C-3TKS65B-21/2/
22b9842f820b08883990bbae1d889c03.
GPBiB
P. J. Angeline. An investigation into the sensitivity of genetic programming to the
frequency of leaf selection during subtree crossover. In J. R. Koza, et al., editors, Genetic Programming 1996: Proceedings of the First Annual Conference, pages
2129, Stanford University, CA, USA, 2831 July 1996. MIT Press. URL http:
//www.natural-selection.com/Library/1996/gp96.zip.
GPBiB
P. J. Angeline. Subtree crossover: Building block engine or macromutation? In J. R.
Koza, et al., editors, Genetic Programming 1997: Proceedings of the Second Annual
Conference, pages 917, Stanford University, CA, USA, 13-16 July 1997. Morgan Kaufmann.
GPBiB
P. J. Angeline and K. E. Kinnear, Jr., editors. Advances in Genetic Programming 2. MIT
Press, Cambridge, MA, USA, 1996. ISBN 0-262-01158-1. URL http://www.cs.bham.
ac.uk/~wbl/aigp2.html.
GPBiB
P. J. Angeline and J. B. Pollack. The evolutionary induction of subroutines. In Proceedings
of the Fourteenth Annual Conference of the Cognitive Science Society, pages 236241,
Bloomington, Indiana, USA, 1992. Lawrence Erlbaum. URL http://www.demo.cs.
brandeis.edu/papers/glib92.pdf.
GPBiB
L. Araujo. Multiobjective genetic programming for natural language parsing and tagging. In T. P. Runarsson, et al., editors, Parallel Problem Solving from Nature PPSN IX, volume 4193 of LNCS, pages 433442, Reykjavik, Iceland, 9-13 September
2006. Springer-Verlag. ISBN 3-540-38990-3. URL http://ppsn2006.raunvis.hi.is/
proceedings/055.pdf.
GPBiB
V. Arkov, C. Evans, P. J. Fleming, D. C. Hill, J. P. Norton, I. Pratt,
D. Rees, and K. Rodriguez-Vazquez.
System identification strategies applied
to aircraft gas turbine engines.
Annual Reviews in Control, 24(1):6781,

BIBLIOGRAPHY

169

2000. URL http://www.sciencedirect.com/science/article/B6V0H-482MDPD-8/2/

dd470648e2228c84efe7e14ca3841b7e.
GPBiB
M. P. Austin, G. Bates, M. A. H. Dempster, V. Leemans, and S. N. Williams.
Adaptive systems for foreign exchange trading. Quantitative Finance, 4(4):37
45, August 2004. ISSN 1469-7688. URL http://www-cfr.jbs.cam.ac.uk/archive/
PRESENTATIONS/seminars/2006/dempster2.pdf.
GPBiB
Y. Azaria and M. Sipper. GP-gammon: Genetically programming backgammon players.
Genetic Programming and Evolvable Machines, 6(3):283300, September 2005a. ISSN
1389-2576. URL http://www.cs.bgu.ac.il/~sipper/papabs/gpgammon.pdf. Published online: 12 August 2005.
GPBiB
Y. Azaria and M. Sipper. Using GP-gammon: Using genetic programming to evolve
backgammon players. In M. Keijzer, et al., editors, Proceedings of the 8th European Conference on Genetic Programming, volume 3447 of Lecture Notes in Computer
Science, pages 132142, Lausanne, Switzerland, 30 March - 1 April 2005b. Springer.
ISBN 3-540-25436-6. URL http://springerlink.metapress.com/openurl.asp?genre=
article&issn=0302-9743&volume=3447&spage=132.
GPBiB
V. Babovic. Emergence, evolution, intelligence; Hydroinformatics - A study of distributed
and decentralised computing using intelligent agents. A. A. Balkema Publishers, Rotterdam, Holland, 1996. ISBN 90-5410-404-X.
GPBiB
M. Bader-El-Den and R. Poli. Generating sat local-search heuristics using a gp hyperheuristic framework. In Proceedings of Evolution Artificielle, October 2007a.
M. B. Bader-El-Den and R. Poli. A GP-based hyper-heuristic framework for evolving
3-SAT heuristics. In D. Thierens, et al., editors, GECCO 07: Proceedings of the 9th
annual conference on Genetic and evolutionary computation, volume 2, pages 1749
1749, London, 7-11 July 2007b. ACM Press. URL http://www.cs.bham.ac.uk/~wbl/
biblio/gecco2007/docs/p1749.pdf.
GPBiB
K. M. S. Badran and P. I. Rockett. The roles of diversity preservation and mutation in
preventing population collapse in multiobjective genetic programming. In D. Thierens,
et al., editors, GECCO 07: Proceedings of the 9th annual conference on Genetic
and evolutionary computation, volume 2, pages 15511558, London, 7-11 July 2007.
ACM Press. URL http://www.cs.bham.ac.uk/~wbl/biblio/gecco2007/docs/p1551.
pdf.
GPBiB
W. Bains, R. Gilbert, L. Sviridenko, J.-M. Gascon, R. Scoffin, K. Birchall, I. Harvey,
and J. Caldwell. Evolutionary computational methods to predict oral bioavailability
QSPRs. Current Opinion in Drug Discovery and Development, 5(1):4451, January
2002.
GPBiB
J. E. Baker. Reducing bias and inefficiency in the selection algorithm. In J. J. Grefenstette,
editor, Proceedings of the Second International Conference on Genetic Algorithms
and their Application, pages 1421, Cambridge, MA, USA, 1987. Lawrence Erlbaum
Associates. ISBN 0-8058-0158-8.
J. Balic. Flexible Manufacturing Systems; Development - Structure - Operation - Handling - Tooling. Manufacturing technology. DAAAM International, Vienna, 1999. ISBN
3-901509-03-8.
GPBiB
S. Baluja and R. Caruana. Removing the genetics from the standard genetic algorithm.
In A. Prieditis and S. Russell, editors, Machine Learning: Proceedings of the Twelfth
International Conference, pages 3846. Morgan Kaufmann Publishers, San Francisco,
CA, 1995.
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=0)

1.5

80
1
100
0.5

60
Fitness

Generation 0
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

170

BIBLIOGRAPHY

W. Banzhaf and W. B. Langdon. Some considerations on the reason for bloat. Genetic
Programming and Evolvable Machines, 3(1):8191, March 2002. ISSN 1389-2576. URL
http://web.cs.mun.ca/~banzhaf/papers/genp_bloat.pdf.
GPBiB
W. Banzhaf. Genetic programming for pedestrians. In S. Forrest, editor, Proceedings
of the 5th International Conference on Genetic Algorithms, ICGA-93, page 628,
University of Illinois at Urbana-Champaign, 17-21 July 1993. Morgan Kaufmann.
URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/ftp.io.com/papers/GenProg_
forPed.ps.Z.
GPBiB
W. Banzhaf, F. D. Francone, and P. Nordin. The effect of extensive use of the mutation
operator on generalization in genetic programming using sparse data sets. In H.-M.
Voigt, et al., editors, Parallel Problem Solving from Nature IV, Proceedings of the
International Conference on Evolutionary Computation, volume 1141 of LNCS, pages
300309, Berlin, Germany, 22-26 September 1996. Springer Verlag. ISBN 3-540-61723X.
GPBiB
W. Banzhaf, P. Nordin, R. E. Keller, and F. D. Francone. Genetic Programming
An Introduction; On the Automatic Evolution of Computer Programs and its Applications. Morgan Kaufmann, San Francisco, CA, USA, January 1998a. ISBN 155860-510-X. URL http://www.elsevier.com/wps/find/bookdescription.cws_home/
677869/description#description.
GPBiB
W. Banzhaf, R. Poli, M. Schoenauer, and T. C. Fogarty, editors. Genetic Programming,
volume 1391 of LNCS, Paris, 14-15 April 1998b. Springer-Verlag. ISBN 3-540-643605.
URL http://www.springer.de/cgi-bin/search_book.pl?isbn=3-540-64360-5.
GPBiB
W. Banzhaf, J. Daida, A. E. Eiben, M. H. Garzon, V. Honavar, M. Jakiela, and R. E.
Smith, editors. GECCO-99: Proceedings of the Genetic and Evolutionary Computation Conference, Orlando, Florida, USA, 13-17 July 1999. Morgan Kaufmann.
ISBN 1-55860-611-4. URL http://www.amazon.com/exec/obidos/ASIN/1558606114/
qid%3D977054373/105-7666192-3217523.
GPBiB
G. J. Barlow. Design of autonomous navigation controllers for unmanned aerial vehicles
using multi-objective genetic programming. Masters thesis, North Carolina State University, Raleigh, NC, USA, March 2004. URL http://www.andrew.cmu.edu/user/gjb/
includes/publications/thesis/barlow2004-thesis/barlow2004-thesis.pdf. GPBiB
S. J. Barrett. Recurring analytical problems within drug discovery and development. In
T. Scheffer and U. Leser, editors, Data Mining and Text Mining for Bioinformatics:
Proceedings of the European Workshop, pages 67, Dubrovnik, Croatia, 22 September 2003. URL http://www2.informatik.hu-berlin.de/~scheffer/publications/
ProceedingsWS2003.pdf. Invited talk.
GPBiB
S. J. Barrett and W. B. Langdon. Advances in the application of machine learning
techniques in drug discovery, design and development. In A. Tiwari, et al., editors,
Applications of Soft Computing: Recent Trends, Advances in Soft Computing, pages
99110, On the World Wide Web, 19 September - 7 October 2005 2006. Springer. ISBN
ISBN 3-540-29123-7. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/
barrett_2005_WSC.pdf.
GPBiB
T. Bartz-Beielstein. Experimental research in evolutionary computation : the new experimentalism. Springer, 2006.

BIBLIOGRAPHY

171

F. H. Bennett, III. Automatic creation of an efficient multi-agent architecture using

genetic programming with architecture-altering operations. In J. R. Koza, et al.,
editors, Genetic Programming 1996: Proceedings of the First Annual Conference,
pages 3038, Stanford University, CA, USA, 2831 July 1996. MIT Press. URL
http://cognet.mit.edu/library/books/view?isbn=0262611279.
GPBiB
F. H. Bennett, III, J. R. Koza, J. Shipman, and O. Stiffelman. Building a parallel computer
system for $18,000 that performs a half peta-flop per day. In W. Banzhaf, et al., editors,
Proceedings of the Genetic and Evolutionary Computation Conference, volume 2, pages
14841490, Orlando, Florida, USA, 13-17 July 1999. Morgan Kaufmann. ISBN 155860-611-4. URL http://www.genetic-programming.com/jkpdf/gecco1999beowulf.
pdf.
GPBiB
H.-G. Beyer, U.-M. OReilly, D. V. Arnold, W. Banzhaf, C. Blum, E. W. Bonabeau,
E. Cantu-Paz, D. Dasgupta, K. Deb, J. A. Foster, E. D. de Jong, H. Lipson, X. Llora,
S. Mancoridis, M. Pelikan, G. R. Raidl, T. Soule, A. M. Tyrrell, J.-P. Watson, and
E. Zitzler, editors. GECCO 2005: Proceedings of the 2005 conference on Genetic
and evolutionary computation, Washington DC, USA, 25-29 June 2005. ACM Press.
ISBN 1-59593-010-8. URL http://portal.acm.org/citation.cfm?id=1068009&jmp=
cit&coll=GUIDE&dl=GUIDE&CFID=48779769&CFTOKEN=55479664#supp.
GPBiB
B. Bhanu, Y. Lin, and K. Krawiec. Evolutionary Synthesis of Pattern Recognition Systems. Monographs in Computer Science. Springer-Verlag, New York, 2005. ISBN
0-387-21295-7. URL http://www.springer.com/west/home/computer/imaging?SGWID=
4-149-22-39144807-detailsPage=ppmmedia|aboutThisBook.
GPBiB
A. S. Bickel and R. W. Bickel. Tree structured rules in genetic algorithms. In J. J. Grefenstette, editor, Genetic Algorithms and their Applications: Proceedings of the second
International Conference on Genetic Algorithms, pages 7781, MIT, Cambridge, MA,
USA, 28-31 July 1987. Lawrence Erlbaum Associates.
GPBiB
S. Bleuler, M. Brack, L. Thiele, and E. Zitzler. Multiobjective genetic programming:
Reducing bloat using SPEA2. In Proceedings of the 2001 Congress on Evolutionary
Computation CEC2001, pages 536543, COEX, World Trade Center, 159 Samseongdong, Gangnam-gu, Seoul, Korea, 27-30 May 2001. IEEE Press. ISBN 0-7803-6658-1.
URL ftp://ftp.tik.ee.ethz.ch/pub/people/zitzler/BBTZ2001b.ps.gz.
GPBiB
T. Blickle. Theory of Evolutionary Algorithms and Application to System Synthesis.
PhD thesis, Swiss Federal Institute of Technology, Zurich, November 1996. URL http:
//www.handshake.de/user/blickle/publications/diss.pdf.
GPBiB
W. Bohm and A. Geyer-Schulz. Exact uniform initialization for genetic programming.
In R. K. Belew and M. Vose, editors, Foundations of Genetic Algorithms IV, pages
379407, University of San Diego, CA, USA, 35 August 1996. Morgan Kaufmann.
ISBN 1-55860-460-X.
GPBiB
C. C. Bojarczuk, H. S. Lopes, and A. A. Freitas. Genetic programming for knowledge discovery in chest-pain diagnosis. IEEE Engineering in Medicine and Biology Magazine,
19(4):3844, July-August 2000. ISSN 0739-5175. URL http://ieeexplore.ieee.org/
iel5/51/18543/00853480.pdf.
GPBiB
P. A. N. Bosman and E. D. de Jong. Grammar transformations in an EDA for genetic
programming. In R. Poli, et al., editors, GECCO 2004 Workshop Proceedings, Seattle,
Washington, USA, 26-30 June 2004a. URL http://www.cs.bham.ac.uk/~wbl/biblio/
gecco2004/WOBU001.pdf.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=1)

1.5

80
1
100
0.5

60
Fitness

Generation 1
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

172

BIBLIOGRAPHY

P. A. N. Bosman and E. D. de Jong. Learning probabilistic tree grammars for genetic

programming. In X. Yao, et al., editors, Parallel Problem Solving from Nature PPSN VIII, volume 3242 of LNCS, pages 192201, Birmingham, UK, 18-22 September
2004b. Springer-Verlag. ISBN 3-540-23092-0. URL http://www.cs.uu.nl/~dejong/
publications/edagpppsn.pdf.
GPBiB
A. Brabazon and M. ONeill. Biologically Inspired Algorithms for Financial Modelling.
Natural Computing Series. Springer, 2006. ISBN 3-540-26252-0.
GPBiB
M. Brameier and W. Banzhaf. A comparison of linear genetic programming and neural
networks in medical data mining. IEEE Transactions on Evolutionary Computation, 5
(1):1726, February 2001. URL http://web.cs.mun.ca/~banzhaf/papers/ieee_taec.
pdf.
GPBiB
M. Brameier and W. Banzhaf. Linear Genetic Programming. Number XVI in Genetic
and Evolutionary Computation. Springer, 2007. ISBN 0-387-31029-0. URL http:
//www.springer.com/west/home/default?SGWID=4-40356-22-173660820-0.
GPBiB
M. Brameier, J. Haan, A. Krings, and R. M. MacCallum. Automatic discovery of
cross-family sequence features associated with protein function. BMC bioinformatics [electronic resource], 7(16), January 12 2006. ISSN 1471-2105. URL http:
//www.biomedcentral.com/content/pdf/1471-2105-7-16.pdf.
GPBiB
S. Brave. Evolving recursive programs for tree search. In P. J. Angeline and K. E.
Kinnear, Jr., editors, Advances in Genetic Programming 2, chapter 10, pages 203220.
MIT Press, Cambridge, MA, USA, 1996. ISBN 0-262-01158-1.
GPBiB
S. Brave and A. S. Wu, editors. Late Breaking Papers at the 1999 Genetic and Evolutionary Computation Conference, Orlando, Florida, USA, 13 July 1999.
GPBiB
M. Brezocnik. Uporaba genetskega programiranja v inteligentnih proizvodnih sistemih.
University of Maribor, Faculty of mechanical engineering, Maribor, Slovenia, 2000.
ISBN 86-435-0306-1. URL http://maja.uni-mb.si/slo/Knjige/2000-03-mon/index.
htm.
GPBiB
M. Brezocnik, J. Balic, and L. Gusel. Artificial intelligence approach to determination
of flow curve. Journal for technology of plasticity, 25(1-2):17, 2000. ISSN 0350-2368.
GPBiB
G. Buason, N. Bergfeldt, and T. Ziemke. Brains, bodies, and beyond: Competitive coevolution of robot controllers, morphologies and environments. Genetic Programming
and Evolvable Machines, 6(1):2551, March 2005. ISSN 1389-2576.
E. K. Burke, M. R. Hyde, and G. Kendall. Evolving bin packing heuristics with genetic
programming. In T. P. Runarsson, et al., editors, Parallel Problem Solving from Nature
- PPSN IX, volume 4193 of LNCS, pages 860869, Reykjavik, Iceland, 9-13 September
2006. Springer-Verlag. ISBN 3-540-38990-3. URL http://www.cs.nott.ac.uk/~mvh/
ppsn2006.pdf.
GPBiB
E. K. Burke, G. Kendall, J. Newall, E. Hart, P. Ross, and S. Schulenburg. Hyperheuristics: an emerging direction in modern search technology. In F. Glover and
G. Kochenberger, editors, Handbook of Metaheuristics, pages 457474. Kluwer Academic Publishers, 2003.
E. K. Burke, M. R. Hyde, G. Kendall, and J. Woodward. Automatic heuristic generation with genetic programming: evolving a jack-of-all-trades or a master of one. In
D. Thierens, et al., editors, GECCO 07: Proceedings of the 9th annual conference on

BIBLIOGRAPHY

173

Genetic and evolutionary computation, volume 2, pages 15591565, London, 7-11 July
2007. ACM Press. URL http://www.cs.bham.ac.uk/~wbl/biblio/gecco2007/docs/
p1559.pdf.
GPBiB
B. F. Buxton, W. B. Langdon, and S. J. Barrett. Data fusion by intelligent classifier
combination. Measurement and Control, 34(8):229234, October 2001. URL http:
//www.cs.ucl.ac.uk/staff/W.Langdon/mc/.
GPBiB
W. Cai, A. Pacheco-Vega, M. Sen, and K. T. Yang. Heat transfer correlations by symbolic
regression. International Journal of Heat and Mass Transfer, 49(23-24):43524359,
November 2006.
GPBiB
E. Cant
u-Paz, J. A. Foster, K. Deb, L. Davis, R. Roy, U.-M. OReilly, H.-G. Beyer, R. K.
Standish, G. Kendall, S. W. Wilson, M. Harman, J. Wegener, D. Dasgupta, M. A.
Potter, A. C. Schultz, K. A. Dowsland, N. Jonoska, and J. F. Miller, editors. Genetic
and Evolutionary Computation GECCO 2003, Part I, volume 2723 of Lecture Notes
in Computer Science, Chicago, IL, USA, 12-16 July 2003. Springer. ISBN 3-540-406026.
GPBiB
F. Castillo, A. Kordon, and G. Smits. Robust pareto front genetic programming parameter selection based on design of experiments and industrial data. In R. L. Riolo, et al.,
editors, Genetic Programming Theory and Practice IV, volume 5 of Genetic and Evolutionary Computation, chapter 2, pages . Springer, Ann Arbor, 11-13 May 2006a.
ISBN 0-387-33375-4.
GPBiB
F. Castillo, A. Kordon, G. Smits, B. Christenson, and D. Dickerson. Pareto front genetic programming parameter selection based on design of experiments and industrial
data. In M. Keijzer, et al., editors, GECCO 2006: Proceedings of the 8th annual
conference on Genetic and evolutionary computation, volume 2, pages 16131620,
Seattle, Washington, USA, 8-12 July 2006b. ACM Press. ISBN 1-59593-186-4. URL
http://www.cs.bham.ac.uk/~wbl/biblio/gecco2006/docs/p1613.pdf.
GPBiB
M. Chami and D. Robilliard. Inversion of oceanic constituents in case I and II waters with
genetic programming algorithms. Applied Optics, 41(30):62606275, October 2002.
URL http://ao.osa.org/ViewMedia.cfm?id=70258&seq=0.
GPBiB
A. Channon. Unbounded evolutionary dynamics in a system of agents that actively process and transform their environment. Genetic Programming and Evolvable Machines,
7(3):253281, October 2006. ISSN 1389-2576.
D. L. Chao and S. Forrest. Information immune systems. Genetic Programming and
Evolvable Machines, 4(4):311331, December 2003. ISSN 1389-2576.
S. M. Cheang, K. S. Leung, and K. H. Lee. Genetic parallel programming: Design and
implementation. Evolutionary Computation, 14(2):129156, Summer 2006. ISSN 10636560.
GPBiB
K. Chellapilla. Evolving computer programs without subtree crossover. IEEE Transactions on Evolutionary Computation, 1(3):209216, September 1997a.
GPBiB
K. Chellapilla. Evolutionary programming with tree mutations: Evolving computer programs without crossover. In J. R. Koza, et al., editors, Genetic Programming 1997:
Proceedings of the Second Annual Conference, pages 431438, Stanford University, CA,
USA, 13-16 July 1997b. Morgan Kaufmann.
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=2)

1.5

80
1
100
0.5

60
Fitness

Generation 2
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

174

BIBLIOGRAPHY

S.-H. Chen, editor.

Genetic Algorithms and Genetic Programming in Computational Finance.
Kluwer Academic Publishers, Dordrecht, July 2002.
ISBN 0-7923-7601-3. URL http://www.springer.com/west/home/business?SGWID=
4-40517-22-33195998-detailsPage=ppmmedia|toc.
GPBiB
S.-H. Chen, J. Duffy, and C.-H. Yeh. Equilibrium selection via adaptation: Using genetic
programming to model learning in a coordination game. The Electronic Journal of
Evolutionary Modeling and Economic Dynamics, 15 January 2002. ISSN 1298-0137.
GPBiB
S.-H. Chen and C.-C. Liao.
Agent-based computational modeling of the
stock price-volume relation.
Information Sciences, 170(1):75100, 18 February
2005. URL http://www.sciencedirect.com/science/article/B6V0C-4B3JHTS-6/2/
9e023835b1c70f176d1903dd3a8b638e.
GPBiB
S.-H. Chen, H.-S. Wang, and B.-T. Zhang. Forecasting high-frequency financial time series
with evolutionary neural trees: The case of heng-sheng stock index. In H. R. Arabnia,
editor, Proceedings of the International Conference on Artificial Intelligence, IC-AI
99, volume 2, pages 437443, Las Vegas, Nevada, USA, 28 June-1 July 1999. CSREA
Press. ISBN 1-892512-17-3. URL http://bi.snu.ac.kr/Publications/Conferences/
International/ICAI99.ps.
GPBiB
D. M. Chitty. A data parallel approach to genetic programming using programmable
graphics hardware. In D. Thierens, et al., editors, GECCO 07: Proceedings of the 9th
annual conference on Genetic and evolutionary computation, volume 2, pages 1566
1573, London, 7-11 July 2007. ACM Press. URL http://www.cs.bham.ac.uk/~wbl/
biblio/gecco2007/docs/p1566.pdf.
GPBiB
S.-B. Cho, N. X. Hoai, and Y. Shan, editors. Proceedings of The First Asian-Pacific
Workshop on Genetic Programming, Rydges (lakeside) Hotel, Canberra, Australia, 8
December 2003. ISBN 0-9751724-0-9.
GPBiB
F. S. Chong and W. B. Langdon. Java based distributed genetic programming on the
internet. In W. Banzhaf, et al., editors, Proceedings of the Genetic and Evolutionary Computation Conference, volume 2, page 1229, Orlando, Florida, USA, 13-17
July 1999. Morgan Kaufmann. ISBN 1-55860-611-4. URL http://www.cs.ucl.ac.uk/
staff/W.Langdon/ftp/papers/p.chong/DGPposter.pdf. Full text in technical report
CSRP-99-7.
GPBiB
V. Ciesielski and X. Li. Analysis of genetic programming runs. In R. I. Mckay and
S.-B. Cho, editors, Proceedings of The Second Asian-Pacific Workshop on Genetic
Programming, Cairns, Australia, 6-7 December 2004. URL http://goanna.cs.rmit.
edu.au/~xiali/pub/ai04.vc.pdf.
GPBiB
R. Cilibrasi, P. Vitanyi, and R. de Wolf. Algorithmic clustering of music based on
string compression. Computer Music Journal, 28(4):4967, Winter 2004. URL
http://homepages.cwi.nl/~paulv/papers/music.pdf.
GPBiB
R. Cilibrasi and P. M. B. Vitanyi. Clustering by compression. IEEE Transactions on
Information Theory, 51(4):15231545, April 2005. URL http://homepages.cwi.nl/
~paulv/papers/cluster.pdf.
GPBiB
J. Clegg, J. A. Walker, and J. F. Miller. A new crossover technique for cartesian genetic
programming. In D. Thierens, et al., editors, GECCO 07: Proceedings of the 9th
annual conference on Genetic and evolutionary computation, volume 2, pages 1580
1587, London, 7-11 July 2007. ACM Press. URL http://www.cs.bham.ac.uk/~wbl/
biblio/gecco2007/docs/p1580.pdf.
GPBiB

BIBLIOGRAPHY

175

P. Collet. Genetic programming. In J.-P. Rennard, editor, Handbook of Research on

Nature-Inspired Computing for Economics and Management, volume I, chapter V,
pages 5973. Idea Group Inc., 1200 E. Colton Ave, 2007. ISBN 1-59140-984-5. GPBiB
P. Collet, M. Tomassini, M. Ebner, S. M. Gustafson, and A. Ek
art, editors. Proceedings
of the 9th European Conference on Genetic Programming, volume 3905 of Lecture
Notes in Computer Science, Budapest, Hungary, 10 - 12 April 2006. Springer. ISBN
3-540-33143-3. URL http://www.springerlink.com/openurl.asp?genre=issue&issn=
0302-9743&volume=3905.
GPBiB
R. J. Collins. Studies in Artificial Evolution. PhD thesis, UCLA, Artificial Life Laboratory, Department of Computer Science, University of California, Los Angeles, LA CA
90024, USA, 1992.
O. Cordon, E. Herrera-Viedma, and M. Luque. Evolutionary learning of boolean queries
by multiobjective genetic programming. In J. J. Merelo-Guervos, et al., editors, Parallel Problem Solving from Nature - PPSN VII, number 2439 in Lecture Notes in Computer Science, LNCS, pages 710719, Granada, Spain, 7-11 September 2002. SpringerVerlag. ISBN 3-540-44139-5. URL http://link.springer.de/link/service/series/
0558/bibs/2439/24390710.htm.
GPBiB
F. Corno, E. Sanchez, and G. Squillero. Evolving assembly programs: how games help
microprocessor validation. Evolutionary Computation, IEEE Transactions on, 9(6):
695706, 2005.
D. Costelloe and C. Ryan. Towards models of user preferences in interactive musical
evolution. In D. Thierens, et al., editors, GECCO 07: Proceedings of the 9th annual conference on Genetic and evolutionary computation, volume 2, pages 22542254,
London, 7-11 July 2007. ACM Press. URL http://www.cs.bham.ac.uk/~wbl/biblio/
gecco2007/docs/p2254.pdf.
GPBiB
N. L. Cramer. A representation for the adaptive generation of simple sequential programs.
In J. J. Grefenstette, editor, Proceedings of an International Conference on Genetic Algorithms and the Applications, pages 183187, Carnegie-Mellon University, Pittsburgh,
PA, USA, 24-26 July 1985. URL http://www.sover.net/~nichael/nlc-publications/
icga85/index.html.
GPBiB
E. F. Crane and N. F. McPhee. The effects of size and depth limits on tree based genetic
programming. In T. Yu, et al., editors, Genetic Programming Theory and Practice III,
volume 9 of Genetic Programming, chapter 15, pages 223240. Springer, Ann Arbor,
12-14 May 2005. ISBN 0-387-28110-X.
GPBiB
R. Crawford-Marks and L. Spector. Size control via size fair genetic operators in the
PushGP genetic programming system. In W. B. Langdon, et al., editors, GECCO
2002: Proceedings of the Genetic and Evolutionary Computation Conference, pages
733739, New York, 9-13 July 2002. Morgan Kaufmann Publishers. ISBN 1-55860-8788. URL http://alum.hampshire.edu/~rpc01/gp234.pdf.
GPBiB
R. L. Crepeau. Genetic evolution of machine language software. In J. P. Rosca, editor,
Proceedings of the Workshop on Genetic Programming: From Theory to Real-World
Applications, pages 121134, Tahoe City, California, USA, 9 July 1995. URL http:
//www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/GEMS_Article.pdf.
GPBiB
R. Curry, P. Lichodzijewski, and M. I. Heywood. Scaling genetic programming to large
datasets using hierarchical dynamic subset selection. IEEE Transactions on Systems,
Man, and Cybernetics: Part B - Cybernetics, 37(4):10651073, August 2007. ISSN
1083-4419. URL http://www.cs.dal.ca/~mheywood/X-files/GradPubs.html#curry.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=3)

1.5

80
1
100
0.5

60
Fitness

Generation 3
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

176

BIBLIOGRAPHY

J. M. Daida, A. M. Hilss, D. J. Ward, and S. L. Long. Visualizing tree structures in

genetic programming. Genetic Programming and Evolvable Machines, 6(1):79110,
March 2005. ISSN 1389-2576.
GPBiB
J. M. Daida, J. D. Hommes, T. F. Bersano-Begey, S. J. Ross, and J. F. Vesecky. Algorithm discovery using the genetic programming paradigm: Extracting low-contrast
curvilinear features from SAR images of arctic ice. In P. J. Angeline and K. E.
Kinnear, Jr., editors, Advances in Genetic Programming 2, chapter 21, pages 417
442. MIT Press, Cambridge, MA, USA, 1996. ISBN 0-262-01158-1. URL http:
//sitemaker.umich.edu/daida/files/GP2_cha21.pdf.
GPBiB
C. Darwin. The Origin of Species. John Murray, penguin classics, 1985 edition, 1859.
ISBN 0-14-043205-1.
E. Dassau, B. Grosman, and D. R. Lewin. Modeling and temperature control of rapid
thermal processing. Computers and Chemical Engineering, 30(4):686697, 15 February
2006. URL http://tx.technion.ac.il/~dlewin/publications/rtp_paper_v9.pdf.
GPBiB
T. E. Davis and J. C. Principe. A Markov chain framework for the simple genetic algorithm. Evolutionary Computation, 1(3):269288, 1993.
J. P. Day, D. B. Kell, and G. W. Griffith. Differentiation of phytophthora infestans
sporangia from other airborne biological particles by flow cytometry. Applied and
Environmental Microbiology, 68(1):3745, January 2002. URL http://intl-aem.asm.
org/cgi/reprint/68/1/37.pdf.
GPBiB
J. S. de Bonet, C. L. Isbell, Jr., and P. Viola. MIMIC: Finding optima by estimating
probability densities. In M. C. M. et. al., editor, Advances in Neural Information
Processing Systems, volume 9, page 424. MIT Press, 1997. URL http://citeseer.
ist.psu.edu/debonet96mimic.html.
E. D. de Jong and J. B. Pollack. Multi-objective methods for tree size control. Genetic
Programming and Evolvable Machines, 4(3):211233, September 2003. ISSN 13892576. URL http://www.cs.uu.nl/~dejong/publications/bloat.ps.
GPBiB
E. D. de Jong, R. A. Watson, and J. B. Pollack. Reducing bloat and promoting diversity
using multi-objective methods. In L. Spector, et al., editors, Proceedings of the Genetic and Evolutionary Computation Conference (GECCO-2001), pages 1118, San
Francisco, California, USA, 7-11 July 2001. Morgan Kaufmann. ISBN 1-55860-774-9.
URL http://www.demo.cs.brandeis.edu/papers/rbpd_gecco01.pdf.
GPBiB
J. S. de Sousa, L. de C. T. Gomes, G. B. Bezerra, L. N. de Castro, and F. J. Von Zuben. An
immune-evolutionary algorithm for multiple rearrangements of gene expression data.
Genetic Programming and Evolvable Machines, 5(2):157179, June 2004. ISSN 13892576.
GPBiB
C. De Stefano, A. D. Cioppa, and A. Marcelli.
Character preclassification
based on genetic programming. Pattern Recognition Letters, 23(12):14391448,
2002. URL http://www.sciencedirect.com/science/article/B6V15-45J91MV-4/2/
3e5c2ac0c51428d0f7ea9fc0142f6790.
GPBiB
K. Deb. Multi-objective optimization using evolutionary algorithms. Wiley, 2001.
K. Deb, S. Agrawal, A. Pratap, and T. Meyarivan. A fast elitist non-dominated sorting
genetic algorithm for multi-objective optimisation: Nsga-ii. In PPSN VI: Proceedings
of the 6th International Conference on Parallel Problem Solving from Nature, pages
849858, London, UK, 2000. Springer-Verlag. ISBN 3-540-41056-2.

BIBLIOGRAPHY

177

K. Deb, R. Poli, W. Banzhaf, H.-G. Beyer, E. Burke, P. Darwen, D. Dasgupta,

D. Floreano, J. Foster, M. Harman, O. Holland, P. L. Lanzi, L. Spector, A. Tettamanzi, D. Thierens, and A. Tyrrell, editors. Genetic and Evolutionary Computation GECCO-2004, Part I, volume 3102 of Lecture Notes in Computer Science,
Seattle, WA, USA, 26-30 June 2004. Springer-Verlag. ISBN 3-540-22344-4. URL
http://www.springerlink.com/content/978-3-540-22344-3.
GPBiB
M. Defoin Platel, M. Clergue, and P. Collard. Maximum homologous crossover for linear
genetic programming. In C. Ryan, et al., editors, Genetic Programming, Proceedings of EuroGP2003, volume 2610 of LNCS, pages 194203, Essex, 14-16 April 2003.
Springer-Verlag. ISBN 3-540-00971-X. URL http://www.i3s.unice.fr/~defoin/
publications/eurogp_03.pdf.
GPBiB
M. A. H. Dempster and C. M. Jones. A real-time adaptive trading system using genetic
programming. Quantitative Finance, 1:397413, 2000. URL http://mahd-pc.jbs.cam.
ac.uk/archive/PAPERS/2000/geneticprogramming.pdf.
GPBiB
M. A. H. Dempster, T. W. Payne, Y. Romahi, and G. W. P. Thompson. Computational
learning techniques for intraday FX trading using popular technical indicators. IEEE
Transactions on Neural Networks, 12(4):744754, July 2001. ISSN 1045-9227. URL
http://mahd-pc.jbs.cam.ac.uk/archive/PAPERS/2000/ieeetrading.pdf.
GPBiB
L. Deschaine. Using information fusion, machine learning, and global optimisation to
increase the accuracy of finding and understanding items interest in the subsurface.
GeoDrilling International, (122):3032, May 2006. URL http://www.mining-journal.
com/gdi_magazine/pdf/GDI0605scr.pdf.
GPBiB
L. M. Deschaine, R. A. Hoover, J. N. Skibinski, J. J. Patel, F. Francone, P. Nordin,
and M. J. Ades. Using machine learning to compliment and extend the accuracy of
UXO discrimination beyond the best reported results of the jefferson proving ground
technology demonstration. In 2002 Advanced Technology Simulation Conference, San
Diego, CA, USA, 14-18 April 2002. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/
ftp/papers/deschaine/ASTC_2002_UXOFinder_Invention_Paper.pdf.
GPBiB
L. M. Deschaine, J. J. Patel, R. D. Guthrie, J. T. Grimski, and M. J. Ades. Using linear
genetic programming to develop a C/C++ simulation model of a waste incinerator.
In M. Ades, editor, Advanced Technology Simulation Conference, Seattle, 22-26 April
2001. URL http://www.aimlearning.com/Environmental.Engineering.pdf. GPBiB
P. Dhaeseleer. Context preserving crossover in genetic programming. In Proceedings of
the 1994 IEEE World Congress on Computational Intelligence, volume 1, pages 256
261, Orlando, Florida, USA, 27-29 June 1994. IEEE Press. URL http://www.cs.ucl.
ac.uk/staff/W.Langdon/ftp/ftp.io.com/papers/WCCI94_CPC.ps.Z.
GPBiB
P. Dhaeseleer and J. Bluming. Effects of locality in individual and population evolution.
In K. E. Kinnear, Jr., editor, Advances in Genetic Programming, chapter 8, pages
177198. MIT Press, 1994. URL http://cognet.mit.edu/library/books/view?isbn=
0262111888.
GPBiB
S. Dignum and R. Poli. Generalisation of the limiting distribution of program sizes in
tree-based genetic programming and analysis of its effects on bloat. In D. Thierens,
et al., editors, GECCO 07: Proceedings of the 9th annual conference on Genetic
and evolutionary computation, volume 2, pages 15881595, London, 7-11 July 2007.
ACM Press. URL http://www.cs.bham.ac.uk/~wbl/biblio/gecco2007/docs/p1588.
pdf.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=4)

1.5

80
1
100
0.5

60
Fitness

Generation 4
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

178

BIBLIOGRAPHY

C. Dimopoulos. A genetic programming methodology for the solution of the multiobjective cell-formation problem. In H.-D. Cheng, editor, Proceedings of the 8th Joint
Conference in Information Systems (JCIS 2005), pages 14871494, Salt Lake City,
USA, 21-25 July 2005.
GPBiB
J. U. Dolinsky, I. D. Jenkinson, and G. J. Colquhoun. Application of genetic programming
to the calibration of industrial robots. Computers in Industry, 58(3):255264, April
2007. ISSN 0166-3615.
GPBiB
R. P. Domingos, R. Schirru, and A. S. Martinez. Soft computing systems applied to
PWRs xenon. Progress in Nuclear Energy, 46(3-4):297308, 2005.
GPBiB
M. Dorigo and T. St
utzle. Ant Colony Optimization. MIT Press (Bradford Books), 2004.
D. C. Dracopoulos. Evolutionary Learning Algorithms for Neural Adaptive Control. Perspectives in Neural Computing. Springer Verlag, P.O. Box 31 13 40, D-10643 Berlin,
Germany, August 1997. ISBN 3-540-76161-6. URL http://www.springer.de/catalog/
html-files/deutsch/comp/3540761616.html.
GPBiB
S. Draves. The electric sheep. SIGEVOlution, 1(2):1016, 2006. ISSN 1931-8499.
S. Droste, T. Jansen, G. Rudolph, H.-P. Schwefel, K. Tinnefeld, and I. Wegener. Theory
of evolutionary algorithms and genetic programming. In H.-P. Schwefel, et al., editors,
Advances in Computational Intelligence: Theory and Practice, Natural Computing
Series, chapter 5, pages 107144. Springer, 2003. ISBN 3-540-43269-8.
GPBiB
M. Ebner, M. Reinhardt, and J. Albert. Evolution of vertex and pixel shaders. In
M. Keijzer, et al., editors, Proceedings of the 8th European Conference on Genetic
Programming, volume 3447 of Lecture Notes in Computer Science, pages 261270,
Lausanne, Switzerland, 30 March - 1 April 2005. Springer. ISBN 3-540-254366.
URL http://springerlink.metapress.com/openurl.asp?genre=article&issn=
0302-9743&volume=3447&spage=261.
GPBiB
M. Ebner, M. ONeill, A. Ek
art, L. Vanneschi, and A. I. Esparcia-Alc
azar, editors. Proceedings of the 10th European Conference on Genetic Programming, volume 4445 of
Lecture Notes in Computer Science, Valencia, Spain, 11 - 13 April 2007. Springer. ISBN
3-540-71602-5. URL http://www.springerlink.com/content/978-3-540-71602-0/.
GPBiB
EC-Digest, 1985-2008. URL http://ec-digest.research.ucf.edu/.
A. E. Eiben and J. E. Smith. Introduction to Evolutionary Computing. Springer, 2003.
ISBN 3-540-40184-9.
GPBiB
A. Ekart and S. Z. Nemeth. Selection based on the pareto nondomination criterion for
controlling code growth in genetic programming. Genetic Programming and Evolvable
Machines, 2(1):6173, March 2001. ISSN 1389-2576.
GPBiB
S. E. Eklund. A massively parallel architecture for linear machine code genetic programming. In Y. Liu, et al., editors, Evolvable Systems: From Biology to Hardware: Proceedings of 4th International Conference, ICES 2001, volume 2210 of
Lecture Notes in Computer Science, pages 216224, Tokyo, Japan, October 35 2001. Springer-Verlag. URL http://www.springerlink.com/openurl.asp?genre=
article&issn=0302-9743&volume=2210&spage=216.
GPBiB
S. E. Eklund. A massively parallel GP engine in VLSI. In D. B. Fogel, et al., editors,
Proceedings of the 2002 Congress on Evolutionary Computation CEC2002, pages 629
633. IEEE Press, 2002. ISBN 0-7803-7278-6.
GPBiB

BIBLIOGRAPHY

179

S. E. Eklund. A massively parallel architecture for distributed genetic algorithms. Parallel

Computing, 30(5-6):647676, 2004. URL http://www.sciencedirect.com/science/
article/B6V12-4CDS49V-1/2/5ba1531eae2c9d8b336f1e90cc0ba5e9.
GPBiB
D. I. Ellis, D. Broadhurst, and R. Goodacre. Rapid and quantitative detection of the microbial spoilage of beef by fourier transform infrared spectroscopy and machine learning. Analytica Chimica Acta, 514(2):193201, 2004. URL http://dbkgroup.org/dave_
files/ACAbeef04.pdf.
GPBiB
D. I. Ellis, D. Broadhurst, D. B. Kell, J. J. Rowland, and R. Goodacre. Rapid and
quantitative detection of the microbial spoilage of meat by fourier transform infrared
spectroscopy and machine learning. Applied and Environmental Microbiology, 68(6):
28222828, June 2002. URL http://dbkgroup.org/Papers/app_%20env_microbiol_
68_(2822).pdf.
GPBiB
R. Eriksson and B. Olsson. Adapting genetic regulatory models by genetic programming.
Biosystems, 76(1-3):217227, 2004. URL http://www.sciencedirect.com/science/
article/B6T2K-4D09KY2-7/2/1abfe196bb4afc60afc3311cadb75d66.
GPBiB
A. I. Esparcia-Alcazar and K. C. Sharman. Genetic programming techniques that evolve
recurrent neural networks architectures for signal processing. In IEEE Workshop on
Neural Networks for Signal Processing, Seiko, Kyoto, Japan, September 1996. GPBiB
C. Evans, P. J. Fleming, D. C. Hill, J. P. Norton, I. Pratt, D. Rees,
and K. Rodriguez-Vazquez.
Application of system identification techniques
to aircraft gas turbine engines.
Control Engineering Practice, 9(2):135148,
2001. URL http://www.sciencedirect.com/science/article/B6V2H-4280YP2-3/1/
24d44180070f91dea854032d98f9187a.
GPBiB
F. Federman, G. Sparkman, and S. Watt. Representation of music in a learning classifier
system utilizing bach chorales. In W. Banzhaf, et al., editors, Proceedings of the
Genetic and Evolutionary Computation Conference, volume 1, page 785, Orlando,
Florida, USA, 13-17 July 1999. Morgan Kaufmann. ISBN 1-55860-611-4.
M. J. Felton. Survival of the fittest in drug design. Modern Drug Discovery, 3(9):4950,
November/December 2000. ISSN 1532-4486. URL http://pubs.acs.org/subscribe/
journals/mdd/v03/i09/html/felton.html.
GPBiB
F. Fernandez, J. M. Sanchez, M. Tomassini, and J. A. Gomez. A parallel genetic programming tool based on PVM. In J. Dongarra, et al., editors, Recent Advances in
Parallel Virtual Machine and Message Passing Interface, Proceedings of the 6th European PVM/MPI Users Group Meeting, volume 1697 of Lecture Notes in Computer
Science, pages 241248, Barcelona, Spain, September 1999. Springer-Verlag. ISBN
3-540-66549-8.
GPBiB
F. Fernandez, M. Tomassini, and L. Vanneschi. An empirical study of multipopulation genetic programming. Genetic Programming and Evolvable Machines, 4(1):2151, March
2003. ISSN 1389-2576.
GPBiB
G. Folino, C. Pizzuti, and G. Spezzano. A scalable cellular implementation of parallel
genetic programming. IEEE Transactions on Evolutionary Computation, 7(1):3753,
February 2003.
GPBiB
J. A. Foster. Review: Discipulus: A commercial genetic programming system. Genetic
Programming and Evolvable Machines, 2(2):201203, June 2001. ISSN 1389-2576.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=5)

1.5

80
1
100
0.5

60
Fitness

Generation 5
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

180

BIBLIOGRAPHY

J. A. Foster, E. Lutton, J. Miller, C. Ryan, and A. G. B. Tettamanzi, editors. Genetic

Programming, Proceedings of the 5th European Conference, EuroGP 2002, volume
2278 of LNCS, Kinsale, Ireland, 3-5 April 2002. Springer-Verlag. ISBN 3-540-43378-3.
GPBiB
F. D. Francone, M. Conrads, W. Banzhaf, and P. Nordin. Homologous crossover in
genetic programming. In W. Banzhaf, et al., editors, Proceedings of the Genetic and
Evolutionary Computation Conference, volume 2, pages 10211026, Orlando, Florida,
USA, 13-17 July 1999. Morgan Kaufmann. ISBN 1-55860-611-4. URL http://www.cs.
bham.ac.uk/~wbl/biblio/gecco1999/GP-463.pdf.
GPBiB
F. D. Francone and L. M. Deschaine. Getting it right at the very start building
project models where data is expensive by combining human expertise, machine learning and information theory. In 2004 Business and Industry Symposium, Washington, DC, April 2004. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/
deschaine/ASTC_2004_Getting_It_Right_from_the_Very_Start.pdf.
GPBiB
A. Fraser and T. Weinbrenner. GPC++ Genetic Programming C++ Class Library, 19931997. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/weinbenner/gp.html.
C. Fujiki and J. Dickinson. Using the genetic algorithm to generate lisp source code
to solve the prisoners dilemma. In J. J. Grefenstette, editor, Genetic Algorithms
and their Applications: Proceedings of the second international conference on Genetic
Algorithms, pages 236240, MIT, Cambridge, MA, USA, 28-31 July 1987. Lawrence
Erlbaum Associates.
A. Fukunaga. Automated discovery of composite SAT variable selection heuristics. In
Proceedings of the National Conference on Artificial Intelligence (AAAI), pages 641
648, 2002.
GPBiB
A. Fukunaga and A. Stechert. Evolving nonlinear predictive models for lossless image
compression with genetic programming. In J. R. Koza, et al., editors, Genetic Programming 1998: Proceedings of the Third Annual Conference, pages 95102, University
of Wisconsin, Madison, Wisconsin, USA, 22-25 July 1998. Morgan Kaufmann. ISBN
1-55860-548-7. URL http://citeseer.ist.psu.edu/507773.html.
GPBiB
A. S. Fukunaga. Evolving local search heuristics for SAT using genetic programming. In
K. Deb, et al., editors, Genetic and Evolutionary Computation GECCO-2004, Part
II, volume 3103 of Lecture Notes in Computer Science, pages 483494, Seattle, WA,
USA, 26-30 June 2004. Springer-Verlag. ISBN 3-540-22343-6. URL http://alexf04.
maclisp.org/gecco2004.pdf.
GPBiB
P. Funes, E. Sklar, H. Juille, and J. Pollack. Animal-animat coevolution: Using the animal
population as fitness function. In R. Pfeifer, et al., editors, From Animals to Animats 5:
Proceedings of the Fifth International Conference on Simulation of Adaptive Behavior,
pages 525533, Zurich, Switzerland, August 17-21 1998a. MIT Press. ISBN 0-26266144-6. URL http://www.demo.cs.brandeis.edu/papers/tronsab98.pdf.
GPBiB
P. Funes, E. Sklar, H. Juille, and J. Pollack. Animal-animat coevolution: Using the animal
population as fitness function. In R. Pfeifer, et al., editors, From Animals to Animats
5: Proceedings of the Fifth International Conference on Simulation of Adaptive Behavior., pages 525533, Zurich, Switzerland, August 17-21 1998b. MIT Press. ISBN
0-262-66144-6. URL http://www.demo.cs.brandeis.edu/papers/tronsab98.html.
C. Gagn
e and M. Parizeau. Open BEAGLE: A new C++ evolutionary computation
framework. In W. B. Langdon, et al., editors, GECCO 2002: Proceedings of the
Genetic and Evolutionary Computation Conference, page 888, New York, 9-13 July

BIBLIOGRAPHY

181

2002. Morgan Kaufmann Publishers. ISBN 1-55860-878-8. URL http://www.cs.bham.

ac.uk/~wbl/biblio/gecco2002/GP272.pdf.
GPBiB
C. Gagne and M. Parizeau. Genetic engineering of hierarchical fuzzy regional representations for handwritten character recognition. International Journal on Document
Analysis and Recognition, 8(4):223231, September 2006. URL http://vision.gel.
ulaval.ca/fr/publications/Id_607/PublDetails.php.
GPBiB
C. Gagn
e and M. Parizeau. Co-evolution of nearest neighbor classifiers. International
Journal of Pattern Recognition and Artificial Intelligence, 21(5):921946, August
2007. ISSN 0218-0014. URL http://vision.gel.ulaval.ca/en/publications/Id_
692/PublDetails.php.
GPBiB
A. L. Garcia-Almanza and E. P. K. Tsang. Simplifying decision trees learned by genetic
programming. In Proceedings of the 2006 IEEE Congress on Evolutionary Computation, pages 79067912, Vancouver, 6-21 July 2006. IEEE Press. ISBN 0-7803-9487-9.
URL http://privatewww.essex.ac.uk/~algarc/Publications/WCCI2006.pdf. GPBiB
A. L. Garcia-Almanza and E. P. K. Tsang. Repository method to suit different investment
strategies. In D. Srinivasan and L. Wang, editors, 2007 IEEE Congress on Evolutionary
Computation, pages 790797, Singapore, 25-28 September 2007. IEEE Computational
Intelligence Society, IEEE Press. ISBN 1-4244-1340-0.
GPBiB
C. Gathercole and P. Ross. Dynamic training subset selection for supervised learning in genetic programming. In Y. Davidor, et al., editors, Parallel Problem Solving from Nature III, volume 866 of LNCS, pages 312321, Jerusalem, 9-14 October
1994. Springer-Verlag. ISBN 3-540-58484-6. URL http://citeseer.ist.psu.edu/
gathercole94dynamic.html.
GPBiB
C. Gathercole and P. Ross. The MAX problem for genetic programming - highlighting
an adverse interaction between the crossover operator and a restriction on tree depth.
Technical report, Department of Artificial Intelligence, University of Edinburgh, 80
South Bridge, Edinburgh, EH1 1HN, UK, 1995. URL http://citeseer.ist.psu.edu/
gathercole95max.html.
GPBiB
C. Gathercole and P. Ross. Tackling the boolean even N parity problem with genetic programming and limited-error fitness. In J. R. Koza, et al., editors, Genetic Programming 1997: Proceedings of the Second Annual Conference, pages 119
127, Stanford University, CA, USA, 13-16 July 1997. Morgan Kaufmann. URL
http://citeseer.ist.psu.edu/79389.html.
GPBiB
S. Gelly, O. Teytaud, N. Bredeche, and M. Schoenauer. Universal consistency and bloat
in GP. Revue dIntelligence Artificielle, 20(6):805827, 2006. ISSN 0992-499X. URL
http://hal.inria.fr/docs/00/11/28/40/PDF/riabloat.pdf. Issue on New Methods
in Machine Learning. Theory and Applications.
GPBiB
Genetic Programming mailing list, 2001-2008. URL http://tech.groups.yahoo.com/
group/genetic_programming/.
R. J. Gilbert, R. Goodacre, A. M. Woodward, and D. B. Kell. Genetic programming: A
novel method for the quantitative analysis of pyrolysis mass spectral data. ANALYTICAL CHEMISTRY, 69(21):43814389, 1997. URL http://pubs.acs.org/journals/
ancham/article.cgi/ancham/1997/69/i21/pdf/ac970460j.pdf.
GPBiB
A. Globus, J. Lawton, and T. Wipke. Automatic molecular design using evolutionary
techniques. In A. Globus and D. Srivastava, editors, The Sixth Foresight Conference
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=6)

1.5

80
1
100
0.5

60
Fitness

Generation 6
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

182

BIBLIOGRAPHY

on Molecular Nanotechnology, Westin Hotel in Santa Clara, CA, USA, November 1215, 1998 1998. URL http://www.foresight.org/Conferences/MNT6/Papers/Globus/
index.html.
GPBiB
D. E. Goldberg. Genetic Algorithms in Search Optimization and Machine Learning.
Addison-Wesley, 1989.
R. Goodacre and R. J. Gilbert. The detection of caffeine in a variety of beverages using
curie-point pyrolysis mass spectrometry and genetic programming. The Analyst, 124:
10691074, 1999.
GPBiB
R. Goodacre. Explanatory analysis of spectroscopic data using machine learning of simple, interpretable rules. Vibrational Spectroscopy, 32(1):3345, 5 August 2003. URL
http://www.biospec.net/learning/Metab06/Goodacre-FTIRmaps.pdf. A collection of
Papers Presented at Shedding New Light on Disease: Optical Diagnostics for the New
Millennium (SPEC 2002) Reims, France 23-27 June 2002.
GPBiB
R. Goodacre, B. Shann, R. J. Gilbert, E. M. Timmins, A. C. McGovern, B. K.
Alsberg, D. B. Kell, and N. A. Logan. The detection of the dipicolinic acid
biomarker in bacillus spores using curie-point pyrolysis mass spectrometry and fouriertransform infrared spectroscopy. Analytical Chemistry, 72(1):119127, 1 January
2000. URL http://pubs.acs.org/cgi-bin/article.cgi/ancham/2000/72/i01/html/
ac990661i.html.
GPBiB
R. Goodacre, S. Vaidyanathan, W. B. Dunn, G. G. Harrigan, and D. B. Kell.
Metabolomics by numbers: acquiring and understanding global metabolite data.
Trends in Biotechnology, 22(5):245252, 1 May 2004. URL http://dbkgroup.org/
Papers/trends%20in%20biotechnology_22_(245).pdf.
GPBiB
F. Gruau. Neural Network Synthesis using Cellular Encoding and the Genetic Algorithm.
PhD thesis, Laboratoire de lInformatique du Parallilisme, Ecole Normale Supirieure de
Lyon, France, 1994. URL ftp://ftp.ens-lyon.fr/pub/LIP/Rapports/PhD/PhD1994/
PhD1994-01-E.ps.Z.
GPBiB
F. Gruau and D. Whitley. Adding learning to the cellular development process: a comparative study. Evolutionary Computation, 1(3):213233, 1993.
GPBiB
F. Gruau. Genetic micro programming of neural networks. In K. E. Kinnear, Jr., editor,
Advances in Genetic Programming, chapter 24, pages 495518. MIT Press, 1994. URL
http://cognet.mit.edu/library/books/view?isbn=0262111888.
GPBiB
F. Gruau. On using syntactic constraints with genetic programming. In P. J. Angeline
and K. E. Kinnear, Jr., editors, Advances in Genetic Programming 2, chapter 19, pages
377394. MIT Press, Cambridge, MA, USA, 1996. ISBN 0-262-01158-1.
GPBiB
S. M. Gustafson. An Analysis of Diversity in Genetic Programming. PhD thesis, School
of Computer Science and Information Technology, University of Nottingham, Nottingham, England, February 2004. URL http://www.cs.nott.ac.uk/~smg/research/
publications/phdthesis-gustafson.pdf.
GPBiB
S. M. Gustafson and E. K. Burke. The speciating island model: An alternative parallel
evolutionary algorithm. Journal of Parallel and Distributed Computing, 66(8):1025
1036, August 2006. Parallel Bioinspired Algorithms.
GPBiB
S. M. Gustafson, E. K. Burke, and N. Krasnogor. On improving genetic programming
for symbolic regression. In D. Corne, et al., editors, Proceedings of the 2005 IEEE
Congress on Evolutionary Computation, volume 1, pages 912919, Edinburgh, UK,
2-5 September 2005. IEEE Press. ISBN 0-7803-9363-5.
GPBiB

BIBLIOGRAPHY

183

R. J. Hampo and K. A. Marko. Application of genetic programming to control of vehicle

systems. In Proceedings of the Intelligent Vehicles 92 Symposium, 1992. June 29 July 1, 1992, Detroit, Mi, USA.
GPBiB
P. Han, S. Zhou, and D. Wang. A multi-objective genetic programming/ NARMAX
approach to chaotic systems identification. In The Sixth World Congress on Intelligent
Control and Automation, WCICA 2006, volume 1, pages 17351739, Dalian, 2006.
IEEE. ISBN 1-4244-0332-4.
GPBiB
S. Handley. On the use of a directed acyclic graph to represent a population of computer
programs. In Proceedings of the 1994 IEEE World Congress on Computational Intelligence, volume 1, pages 154159, Orlando, Florida, USA, 27-29 June 1994. IEEE
Press.
GPBiB
S. Handley. Automatic learning of a detector for alpha-helices in protein sequences via
genetic programming. In S. Forrest, editor, Proceedings of the 5th International Conference on Genetic Algorithms, ICGA-93, pages 271278, University of Illinois at UrbanaChampaign, 17-21 July 1993. Morgan Kaufmann.
GPBiB
J. V. Hansen. Genetic programming experiments with standard and homologous crossover
methods. Genetic Programming and Evolvable Machines, 4(1):5366, March 2003.
ISSN 1389-2576.
GPBiB
J. V. Hansen, P. B. Lowry, R. D. Meservy, and D. M. McDonald. Genetic programming
for prevention of cyberterrorism through dynamic and evolving intrusion detection. Decision Support Systems, 43(4):13621374, August 2007. Special Issue Clusters. GPBiB
S. Harding and W. Banzhaf. Fast genetic programming on GPUs. In M. Ebner, et al.,
editors, Proceedings of the 10th European Conference on Genetic Programming, volume 4445 of Lecture Notes in Computer Science, pages 90101, Valencia, Spain, 11 13 April 2007. Springer. ISBN 3-540-71602-5.
GPBiB
G. Harik. Linkage learning via probabilistic modeling in the ECGA. IlliGAL Report
99010, University of Illinois at Urbana-Champaign, 1999.
K. Harries and P. Smith. Exploring alternative operators and search strategies in genetic
programming. In J. R. Koza, et al., editors, Genetic Programming 1997: Proceedings
of the Second Annual Conference, pages 147155, Stanford University, CA, USA, 1316 July 1997. Morgan Kaufmann. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/
ftp/papers/harries.gp97_paper.ps.gz.
GPBiB
G. G. Harrigan, R. H. LaPlante, G. N. Cosma, G. Cockerell, R. Goodacre, J. F. Maddox,
J. P. Luyendyk, P. E. Ganey, and R. A. Roth. Application of high-throughput fouriertransform infrared spectroscopy in toxicology studies: contribution to a study on the
development of an animal model for idiosyncratic toxicity. Toxicology Letters, 146(3):
197205, 2 February 2004.
GPBiB
C. Harris and B. Buxton. GP-COM: A distributed, component-based genetic programming system in C++. In J. R. Koza, et al., editors, Genetic Programming 1996:
Proceedings of the First Annual Conference, page 425, Stanford University, CA, USA,
2831 July 1996. MIT Press. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/
papers/gp96com.ps.gz.
GPBiB
B. Harvey, J. Foster, and D. Frincke. Towards byte code genetic programming. In
W. Banzhaf, et al., editors, Proceedings of the Genetic and Evolutionary Computation
Conference, volume 2, page 1234, Orlando, Florida, USA, 13-17 July 1999. Morgan
Kaufmann. ISBN 1-55860-611-4. URL http://citeseer.ist.psu.edu/468509.html.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=7)

1.5

80
1
100
0.5

60
Fitness

Generation 7
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

184

BIBLIOGRAPHY

S. Hasan, S. Daugelat, P. S. S. Rao, and M. Schreiber. Prioritizing genomic drug targets in

pathogens: Application to mycobacterium tuberculosis. PLoS Computational Biology,
2(6):e61, June 2006. URL http://compbiol.plosjournals.org/archive/1553-7358/
2/6/pdf/10.1371_journal.pcbi.0020061-L.pdf.
T. Hastie, R. Tibshirani, and J. Friedman. The elements of statistical learning: Data
mining, inference, and prediction. Springer, 2001.
A. Hauptman and M. Sipper. GP-endchess: Using genetic programming to evolve chess
endgame players. In M. Keijzer, et al., editors, Proceedings of the 8th European Conference on Genetic Programming, volume 3447 of Lecture Notes in Computer Science,
pages 120131, Lausanne, Switzerland, 30 March - 1 April 2005. Springer. ISBN 3-54025436-6. URL http://www.cs.bgu.ac.il/~sipper/papabs/eurogpchess-final.pdf.
GPBiB
A. Hauptman and M. Sipper. Evolution of an efficient search algorithm for the mate-in-N
problem in chess. In M. Ebner, et al., editors, Proceedings of the 10th European Conference on Genetic Programming, volume 4445 of Lecture Notes in Computer Science,
pages 7889, Valencia, Spain, 11 - 13 April 2007. Springer. ISBN 3-540-71602-5. GPBiB
T. Haynes, R. Wainwright, S. Sen, and D. Schoenefeld. Strongly typed genetic programming in evolving cooperation strategies. In L. Eshelman, editor, Genetic Algorithms:
Proceedings of the Sixth International Conference (ICGA95), pages 271278, Pittsburgh, PA, USA, 15-19 July 1995. Morgan Kaufmann. ISBN 1-55860-370-0. URL
http://www.mcs.utulsa.edu/~rogerw/papers/Haynes-icga95.pdf.
GPBiB
T. D. Haynes, D. A. Schoenefeld, and R. L. Wainwright. Type inheritance in strongly
typed genetic programming. In P. J. Angeline and K. E. Kinnear, Jr., editors, Advances
in Genetic Programming 2, chapter 18, pages 359376. MIT Press, Cambridge, MA,
USA, 1996. ISBN 0-262-01158-1. URL http://www.mcs.utulsa.edu/~rogerw/papers/
Haynes-hier.pdf.
GPBiB
J. He, X. Wang, M. Zhang, J. Wang, and Q. Fang. New research on scalability of
lossless image compression by GP engine. In J. Lohn, et al., editors, Proceedings of
the 2005 NASA/DoD Conference on Evolvable Hardware, pages 160164, Washington,
DC, USA, 29 June-1 July 2005. IEEE Press. ISBN 0-7695-2399-4. URL http://doi.
ieeecomputersociety.org/10.1109/EH.2005.35.
GPBiB
A. G. Heidema, J. M. A. Boer, N. Nagelkerke, E. C. M. Mariman, D. L. van der A, and
E. J. M. Feskens. The challenge for genetic epidemiologists: how to analyze large numbers of SNPs in relation to complex diseases. BMC Genetics, 7(23), April 21 2006. ISSN
1471-2156. URL http://www.biomedcentral.com/content/pdf/1471-2156-7-23.pdf.
GPBiB
M. I. Heywood and A. N. Zincir-Heywood. Register based genetic programming on FPGA
computing platforms. In R. Poli, et al., editors, Genetic Programming, Proceedings
of EuroGP2000, volume 1802 of LNCS, pages 4459, Edinburgh, 15-16 April 2000.
Springer-Verlag. ISBN 3-540-67339-3. URL http://www.springerlink.com/openurl.
asp?genre=article&issn=0302-9743&volume=1802&spage=44.
GPBiB
W. D. Hillis. Co-evolving parasites improve simulated evolution as an optimization procedure. In C. G. Langton, et al., editors, Artificial Life II, volume X of Santa Fe
Institute Studies in the Sciences of Complexity, pages 313324. Addison-Wesley, Santa
Fe Institute, New Mexico, USA, February 1990 1992.

BIBLIOGRAPHY

185

M. Hinchliffe, M. Willis, and M. Tham. Chemical process sytems modelling using multiobjective genetic programming. In J. R. Koza, et al., editors, Genetic Programming
1998: Proceedings of the Third Annual Conference, pages 134139, University of Wisconsin, Madison, Wisconsin, USA, 22-25 July 1998. Morgan Kaufmann. ISBN 1-55860548-7.
GPBiB
M. P. Hinchliffe and M. J. Willis.
Dynamic systems modelling using genetic programming.
Computers & Chemical Engineering, 27(12):18411854,
2003. URL http://www.sciencedirect.com/science/article/B6TFT-49MDYGW-2/2/
742bcc7f22240c7a0381027aa5ff7e73.
GPBiB
S.-Y. Ho, C.-H. Hsieh, H.-M. Chen, and H.-L. Huang. Interpretable gene expression
classifier with an accurate and compact fuzzy rule base for microarray data analysis.
Biosystems, 85(3):165176, September 2006.
GPBiB
N. X. Hoai and R. I. McKay. Softening the structural difficulty in genetic programming
with TAG-based representation and insertion/deletion operators. In K. Deb, et al.,
editors, Genetic and Evolutionary Computation GECCO-2004, Part II, volume 3103
of Lecture Notes in Computer Science, pages 605616, Seattle, WA, USA, 26-30 June
2004. Springer-Verlag. ISBN 3-540-22343-6. URL http://link.springer.de/link/
service/series/0558/bibs/3103/31030605.htm.
GPBiB
N. X. Hoai, R. I. McKay, and H. A. Abbass. Tree adjoining grammars, language bias,
and genetic programming. In C. Ryan, et al., editors, Genetic Programming, Proceedings of EuroGP2003, volume 2610 of LNCS, pages 335344, Essex, 14-16 April 2003.
Springer-Verlag. ISBN 3-540-00971-X. URL http://www.cs.adfa.edu.au/~abbass/
publications/hardcopies/TAG3P-EuroGp-03.pdf.
GPBiB
N. X. Hoai, R. I. B. McKay, and D. Essam. Representation and structural difficulty in
genetic programming. IEEE Transactions on Evolutionary Computation, 10(2):157
166, April 2006. URL http://sc.snu.ac.kr/courses/2006/fall/pg/aai/GP/nguyen/
Structdiff.pdf.
GPBiB
N. X. Hoai, R. I. McKay, D. Essam, and H. T. Hao. Genetic transposition in treeadjoining grammar guided genetic programming: The duplication operator. In
M. Keijzer, et al., editors, Proceedings of the 8th European Conference on Genetic
Programming, volume 3447 of Lecture Notes in Computer Science, pages 108119,
Lausanne, Switzerland, 30 March - 1 April 2005. Springer. ISBN 3-540-254366.
URL http://springerlink.metapress.com/openurl.asp?genre=article&issn=
0302-9743&volume=3447&spage=108.
GPBiB
T.-H. Hoang, D. Essam, R. I. B. McKay, and X. H. Nguyen. Building on success in genetic
programming:adaptive variation & developmental evaluation. In Proceedings of the
2007 International Symposium on Intelligent Computation and Applications (ISICA),
Wuhan, China, September 21-23 2007. China University of Geosciences Press. URL
http://sc.snu.ac.kr/PAPERS/dtag.pdf.
GPBiB
J. H. Holland. Adaptation in Natural and Artificial Systems: An Introductory Analysis
with Applications to Biology, Control and Artificial Intelligence. MIT Press, 1992.
First Published by University of Michigan Press 1975.
P. Holmes. The odin genetic programming system. Tech Report RR-95-3, Computer
Studies, Napier University, Craiglockhart, 216 Colinton Road, Edinburgh, EH14 1DJ,
1995. URL http://citeseer.ist.psu.edu/holmes95odin.html.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=8)

1.5

80
1
100
0.5

60
Fitness

Generation 8
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

186

BIBLIOGRAPHY

P. Holmes and P. J. Barclay. Functional languages on linear chromosomes. In J. R.

Koza, et al., editors, Genetic Programming 1996: Proceedings of the First Annual
Conference, page 427, Stanford University, CA, USA, 2831 July 1996. MIT Press.
GPBiB
J.-H. Hong and S.-B. Cho. The classification of cancer based on DNA microarray data
that uses diverse ensemble genetic programming. Artificial Intelligence In Medicine,
36(1):4358, January 2006.
GPBiB
G. S. Hornby and J. B. Pollack. Body-brain co-evolution using L-systems as a generative
encoding. In L. Spector, et al., editors, Proceedings of the Genetic and Evolutionary
Computation Conference (GECCO-2001), pages 868875, San Francisco, California,
USA, 7-11 July 2001. Morgan Kaufmann. ISBN 1-55860-774-9. URL http://www.
demo.cs.brandeis.edu/papers/hornby_gecco01.pdf.
GPBiB
D. Howard and S. C. Roberts. Incident detection on highways. In U.-M. OReilly, et al.,
editors, Genetic Programming Theory and Practice II, chapter 16, pages 263282.
Springer, Ann Arbor, 13-15 May 2004. ISBN 0-387-23253-2.
GPBiB
D. Howard, S. C. Roberts, and R. Brankin.
Target detection in imagery
by genetic programming.
Advances in Engineering Software, 30(5):303311,
1999. URL http://www.sciencedirect.com/science/article/B6V1P-3W1XV4H-1/1/
6e7aee809f33757d0326c62a21824411.
GPBiB
D. Howard, S. C. Roberts, and C. Ryan. Pragmatic genetic programming strategy for the
problem of vehicle detection in airborne reconnaissance. Pattern Recognition Letters,
27(11):12751288, August 2006. Evolutionary Computer Vision and Image Understanding.
GPBiB
W. H. Hsu and S. M. Gustafson. Wrappers for automatic parameter tuning in multiagent optimization by genetic programming. In IJCAI-2001 Workshop on Wrappers
for Performance Enhancement in Knowledge Discovery in Databases (KDD), Seattle,
Washington, USA, 4 August 2001.
GPBiB
H. Iba. Random tree generation for genetic programming. In H.-M. Voigt, et al., editors,
Parallel Problem Solving from Nature IV, Proceedings of the International Conference
on Evolutionary Computation, volume 1141 of LNCS, pages 144153, Berlin, Germany,
22-26 September 1996a. Springer Verlag. ISBN 3-540-61723-X.
GPBiB
H. Iba. Genetic Programming. Tokyo Denki University Press, 1996b.
H.

GPBiB

Iba.
Complexity-based fitness evaluation for variable length representation.
Position paper at the Workshop on Evolutionary Computation
with Variable Size Representation at ICGA-97, 20 July 1997.
URL http:
//coblitz.codeen.org:3125/citeseer.ist.psu.edu/cache/papers/cs/16452/http:
zSzzSzwww.miv.t.u-tokyo.ac.jpzSz~ibazSztmpzSzagp94.pdf/iba94genetic.pdf.
GPBiB

H. Iba, H. de Garis, and T. Sato. Genetic programming using a minimum description

length principle. In K. E. Kinnear, Jr., editor, Advances in Genetic Programming,
chapter 12, pages 265284. MIT Press, 1994. URL http://citeseer.ist.psu.edu/
327857.html.
GPBiB
H. Iba, H. de Garis, and T. Sato. Temporal data processing using genetic programming.
In L. Eshelman, editor, Genetic Algorithms: Proceedings of the Sixth International
Conference (ICGA95), pages 279286, Pittsburgh, PA, USA, 15-19 July 1995a. Morgan
Kaufmann. ISBN 1-55860-370-0.
GPBiB

BIBLIOGRAPHY

187

H. Iba, T. Sato, and H. de Garis. Recombination guidance for numerical genetic programming. In 1995 IEEE Conference on Evolutionary Computation, volume 1, pages
97102, Perth, Australia, 29 November - 1 December 1995b. IEEE Press.
GPBiB
Y. Inagaki. On synchronized evolution of the network of automata. IEEE Transactions on Evolutionary Computation, 6(2):147158, April 2002. ISSN 1089-778X.
URL http://ieeexplore.ieee.org/iel5/4235/21497/00996014.pdf?tp=&arnumber=
996014&isnumber=21497&arSt=147&ared=158&arAuthor=Inagaki%2C+Y.%3B.
GPBiB
C. Jacob. Principia Evolvica Simulierte Evolution mit Mathematica. dpunkt.verlag,
Heidelberg, Germany, August 1997. ISBN 3-920993-48-9.
GPBiB
C. Jacob. The art of genetic programming. IEEE Intelligent Systems, 15(3):8384, MayJune 2000. ISSN 1094-7167. URL http://ieeexplore.ieee.org/iel5/5254/18363/
00846288.pdf.
GPBiB
C. Jacob. Illustrating Evolutionary Computation with Mathematica. Morgan Kaufmann,
2001. ISBN 1-55860-637-8. URL http://www.mkp.com/books_catalog/catalog.asp?
ISBN=1-55860-637-8.
GPBiB
N. Jin. Equilibrium selection by co-evolution for bargaining problems under incomplete
information about time preferences. In Proceedings of the IEEE Congress on Evolutionary Computation (CEC), pages 26612668, Edinburgh, 25 September 2005.
N. Jin and E. P. K. Tsang. Co-adaptive strategies for sequential bargaining problems with
discount factors and outside options. In Proceedings of the 2006 IEEE Congress on
Evolutionary Computation, pages 79137920, Vancouver, 6-21 July 2006. IEEE Press.
ISBN 0-7803-9487-9.
GPBiB
H. E. Johnson, R. J. Gilbert, M. K. Winson, R. Goodacre, A. R. Smith, J. J. Rowland,
M. A. Hall, and D. B. Kell. Explanatory analysis of the metabolome using genetic
programming of simple, interpretable rules. Genetic Programming and Evolvable Machines, 1(3):243258, July 2000. ISSN 1389-2576.
GPBiB
A. Jones, D. Young, J. Taylor, D. B. Kell, and J. J. Rowland. Quantification of microbial
productivity via multi-angle light scattering and supervised learning. Biotechnology
and Bioengineering, 59(2):131143, 20 July 1998. ISSN 0006-3592.
GPBiB
Jong-Wan Kim. Proceedings of the 2001 Congress on Evolutionary Computation
CEC2001, COEX, World Trade Center, 159 Samseong-dong, Gangnam-gu, Seoul, Korea, 27-30 May 2001. IEEE Press. ISBN 0-7803-6658-1.
GPBiB
E. Jordaan, J. den Doelder, and G. Smits. Novel approach to develop structure-property
relationships using genetic programming. In T. P. Runarsson, et al., editors, Parallel Problem Solving from Nature - PPSN IX, volume 4193 of LNCS, pages 322331,
Reykjavik, Iceland, 9-13 September 2006. Springer-Verlag. ISBN 3-540-38990-3. GPBiB
E. Jordaan, A. Kordon, L. Chiang, and G. Smits. Robust inferential sensors based on
ensemble of predictors generated by genetic programming. In X. Yao, et al., editors,
Parallel Problem Solving from Nature - PPSN VIII, volume 3242 of LNCS, pages
522531, Birmingham, UK, 18-22 September 2004. Springer-Verlag. ISBN 3-54023092-0.
URL http://www.springerlink.com/openurl.asp?genre=article&issn=
0302-9743&volume=3242&spage=522.
GPBiB
A. K. Joshi and Y. Schabes. Tree adjoining grammars. In G. Rozenber and A. Saloma,
editors, Handbook of of Formal Languages, pages 69123. Springer-Verlag, 1997.
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=9)

1.5

80
1
100
0.5

60
Fitness

Generation 9
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

188

BIBLIOGRAPHY

H. Juille and J. B. Pollack. Massively parallel genetic programming. In P. J. Angeline

and K. E. Kinnear, Jr., editors, Advances in Genetic Programming 2, chapter 17,
pages 339358. MIT Press, Cambridge, MA, USA, 1996. ISBN 0-262-01158-1. URL
http://www.demo.cs.brandeis.edu/papers/gp2.pdf.
GPBiB
M. Kaboudan. A measure of time series predictability using genetic programming applied
to stock returns. Journal of Forecasting, 18:345357, 1999.
GPBiB
M. A. Kaboudan. Genetic programming prediction of stock prices. Computational Economics, 6(3):207236, December 2000.
GPBiB
M. Kaboudan. Extended daily exchange rates forecasts using wavelet temporal resolutions. New Mathematics and Natural Computing, 1:79107, 2005.
GPBiB
T. Kalganova and J. Miller. Evolving more efficient digital circuits by allowing circuit
layout evolution and multi-objective fitness. In A. Stoica, et al., editors, The First
NASA/DoD Workshop on Evolvable Hardware, pages 5463, Pasadena, California,
19-21 July 1999. IEEE Computer Society. ISBN 0-7695-0256-3.
GPBiB
A. Kattan and R. Poli. Evolutionary lossless compression with GP-ZIP. In Proceedings
of the IEEE World Congress on Computational Intelligence, 2008.
C. Keber and M. G. Schuster. Option valuation with generalized ant programming. In
W. B. Langdon, et al., editors, GECCO 2002: Proceedings of the Genetic and Evolutionary Computation Conference, pages 7481, New York, 9-13 July 2002. Morgan
Kaufmann Publishers. ISBN 1-55860-878-8. URL http://www.cs.bham.ac.uk/~wbl/
biblio/gecco2002/aaaa075.ps.
GPBiB
M. Keijzer. Efficiently representing populations in genetic programming. In P. J. Angeline
and K. E. Kinnear, Jr., editors, Advances in Genetic Programming 2, chapter 13, pages
259278. MIT Press, Cambridge, MA, USA, 1996. ISBN 0-262-01158-1.
GPBiB
M. Keijzer. Scaled symbolic regression. Genetic Programming and Evolvable Machines,
5(3):259269, September 2004. ISSN 1389-2576.
GPBiB
M. Keijzer, U.-M. OReilly, S. M. Lucas, E. Costa, and T. Soule, editors. Genetic Programming 7th European Conference, EuroGP 2004, Proceedings, volume
3003 of LNCS, Coimbra, Portugal, 5-7 April 2004. Springer-Verlag.
ISBN 3540-21346-5. URL http://www.springerlink.com/openurl.asp?genre=issue&issn=
0302-9743&volume=3003.
GPBiB
M. Keijzer, A. Tettamanzi, P. Collet, J. I. van Hemert, and M. Tomassini, editors. Proceedings of the 8th European Conference on Genetic Programming, volume 3447 of
Lecture Notes in Computer Science, Lausanne, Switzerland, 30 March - 1 April 2005.
Springer. ISBN 3-540-25436-6. URL http://www.springerlink.com/openurl.asp?
genre=issue&issn=0302-9743&volume=3447.
GPBiB
M. Keijzer, M. Cattolico, D. Arnold, V. Babovic, C. Blum, P. Bosman, M. V. Butz,
C. Coello Coello, D. Dasgupta, S. G. Ficici, J. Foster, A. Hernandez-Aguirre,
G. Hornby, H. Lipson, P. McMinn, J. Moore, G. Raidl, F. Rothlauf, C. Ryan, and
D. Thierens, editors. GECCO 2006: Proceedings of the 8th annual conference on Genetic and evolutionary computation, Seattle, Washington, USA, 8-12 July 2006. ACM
Press. ISBN 1-59593-010-8. URL http://portal.acm.org/citation.cfm?id=1143997.
GPBiB
D. B. Kell. Defence against the flood. Bioinformatics World, pages 1618, January/February 2002a. URL http://dbkgroup.org/Papers/biwpp16-18_as_publ.pdf.
GPBiB

BIBLIOGRAPHY

189

D. B. Kell. Metabolomics and machine learning: Explanatory analysis of complex

metabolome data using genetic programming to produce simple, robust rules. Molecular Biology Reports, 29(1-2):237241, 2002b. URL http://dbkgroup.org/Papers/
btk2002_dbk.pdf.
GPBiB
D. B. Kell. Genotype-phenotype mapping: genes as computer programs. Trends in Genetics, 18(11):555559, November 2002c. URL http://dbkgroup.org/Papers/trends_
genet_18_(555).pdf.
GPBiB
D. B. Kell, R. M. Darby, and J. Draper. Genomic computing. explanatory analysis of
plant expression profiling data using machine learning. Plant Physiology, 126(3):943
951, July 2001. URL http://www.plantphysiol.org/cgi/content/full/126/3/943.
GPBiB
R. E. Keller and R. Poli. Linear genetic programming of metaheuristics. In D. Thierens,
et al., editors, GECCO 07: Proceedings of the 9th annual conference on Genetic
and evolutionary computation, volume 2, pages 17531753, London, 7-11 July 2007a.
ACM Press. URL http://www.cs.bham.ac.uk/~wbl/biblio/gecco2007/docs/p1753.
pdf.
GPBiB
R. E. Keller and R. Poli. Cost-benefit investigation of a genetic-programming hyperheuristic. In Proceedings of Evolution Artificielle, October 2007b.
R. E. Keller and R. Poli. Linear genetic programming of parsimonious metaheuristics.
In Proceedings of IEEE Congress on Evolutionary Computation (CEC), September
2007c.
D. Keymeulen, A. Stoica, J. Lohn, and R. S. Zebulum, editors. The Third NASA/DoD
workshop on Evolvable Hardware, Long Beach, California, 12-14 July 2001. IEEE Computer Society. ISBN 0-7695-1180-5. URL http://cism.jpl.nasa.gov/ehw/events/
nasaeh01/.
A. Khan. Intelligent Perceptual Shaping of a Digital Watermark. PhD thesis, Computer Science and Engineering, Ghulam Ishaq Khan Institute of Engineering Sciences
and Technology, Topi, Pakistan, May 2006. URL http://www.cs.ucl.ac.uk/staff/W.
Langdon/ftp/papers/Intelligent_perceptual_shaping_WM_asif.pdf.
GPBiB
A. Khan and A. M. Mirza. Genetic perceptual shaping: Utilizing cover image and conceivable attack information during watermark embedding. Information Fusion, 8(4):
354365, October 2007. ISSN 1566-2535.
GPBiB
B. KHosraviani. Organization design optimization using genetic programming. In J. R.
Koza, editor, Genetic Algorithms and Genetic Programming at Stanford 2003, pages
109117. Stanford Bookstore, Stanford, California, 94305-3079 USA, 4 December 2003.
URL http://www.genetic-programming.org/sp2003/KHosraviani.pdf.
GPBiB
B. KHosraviani, R. E. Levitt, and J. R. Koza. Organization design optimization using
genetic programming. In M. Keijzer, editor, Late Breaking Papers at the 2004 Genetic
and Evolutionary Computation Conference, Seattle, Washington, USA, 26 July 2004.
URL http://www.cs.bham.ac.uk/~wbl/biblio/gecco2004/LBP056.pdf.
GPBiB
R. H. Kibria and Y. Li. Optimizing the initialization of dynamic decision heuristics
in DPLL SAT solvers using genetic programming. In P. Collet, et al., editors, Proceedings of the 9th European Conference on Genetic Programming, volume 3905 of
Lecture Notes in Computer Science, pages 331340, Budapest, Hungary, 10 - 12 April
2006. Springer. ISBN 3-540-33143-3. URL http://link.springer.de/link/service/
series/0558/papers/3905/39050331.pdf.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=11)

1.5

80
1
100
0.5

60
Fitness

Generation 11
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

190

BIBLIOGRAPHY

K. E. Kinnear, Jr. Evolving a sort: Lessons in genetic programming. In Proceedings

of the 1993 International Conference on Neural Networks, volume 2, pages 881
888, San Francisco, USA, 28 March-1 April 1993. IEEE Press. ISBN 0-7803-0999-5.
URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/ftp.io.com/papers/kinnear.
icnn93.ps.Z.
GPBiB
K. E. Kinnear, Jr. Fitness landscapes and difficulty in genetic programming. In Proceedings of the 1994 IEEE World Conference on Computational Intelligence, volume 1,
pages 142147, Orlando, Florida, USA, 27-29 June 1994a. IEEE Press. ISBN 0-78031899-4. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/ftp.io.com/papers/
kinnear.wcci.ps.Z.
GPBiB
K. E. Kinnear, Jr. A perspective on the work in this book. In K. E. Kinnear, Jr., editor,
Advances in Genetic Programming, chapter 1, pages 319. MIT Press, 1994b. URL
http://cognet.mit.edu/library/books/view?isbn=0262111888.
GPBiB
K. E. Kinnear, Jr., editor. Advances in Genetic Programming. MIT Press, Cambridge,
MA, 1994c. URL http://mitpress.mit.edu/book-home.tcl?isbn=0262111888. GPBiB
A. Klappenecker and F. U. May. Evolving better wavelet compression schemes. In A. F.
Laine, et al., editors, Wavelet Applications in Signal and Image Processing III, volume
2569, San Diego, CA, USA, 9-14 July 1995. SPIE. ISBN 0-8194-1928-1. URL http:
//citeseer.ist.psu.edu/334994.html.
GPBiB
T. J. Klassen and M. I. Heywood. Towards the on-line recognition of arabic characters. In Proceedings of the 2002 International Joint Conference on Neural Networks
IJCNN02, pages 19001905, Hilton Hawaiian Village Hotel, Honolulu, Hawaii, 1217 May 2002. IEEE Press. ISBN 0-7803-7278-6. URL http://users.cs.dal.ca/
~mheywood/X-files/Publications/IEEEarabic.pdf.
GPBiB
J. Klein and L. Spector. Unwitting distributed genetic programming via asynchronous
javascript and XML. In D. Thierens, et al., editors, GECCO 07: Proceedings of
the 9th annual conference on Genetic and evolutionary computation, volume 2, pages
16281635, London, 7-11 July 2007. ACM Press. URL http://www.cs.bham.ac.uk/
~wbl/biblio/gecco2007/docs/p1628.pdf.
GPBiB
A. Kordon. Evolutionary computation at dow chemical. SIGEVOlution, 1(3):49, September 2006. URL http://www.sigevolution.org/2006/03/issue.pdf.
GPBiB
A. Kordon, F. Castillo, G. Smits, and M. Kotanchek. Application issues of genetic programming in industry. In T. Yu, et al., editors, Genetic Programming Theory and
Practice III, volume 9 of Genetic Programming, chapter 16, pages 241258. Springer,
Ann Arbor, 12-14 May 2005. ISBN 0-387-28110-X.
GPBiB
M. Kotanchek, G. Smits, and E. Vladislavleva. Pursuing the pareto paradigm tournaments, algorithm variations & ordinal optimization. In R. L. Riolo, et al., editors,
Genetic Programming Theory and Practice IV, volume 5 of Genetic and Evolutionary Computation, chapter 3, pages . Springer, Ann Arbor, 11-13 May 2006. ISBN
0-387-33375-4.
GPBiB
M. Kovacic and J. Balic. Evolutionary programming of a CNC cutting machine. International journal for advanced manufacturing technology, 22(1-2):118124, September 2003. ISSN 0268-3768. URL http://www.springerlink.com/openurl.asp?genre=
article&eissn=1433-3015&volume=22&issue=1&spage=118.
GPBiB

BIBLIOGRAPHY

191

J. R. Koza. Hierarchical genetic algorithms operating on populations of computer

programs. In N. S. Sridharan, editor, Proceedings of the Eleventh International
Joint Conference on Artificial Intelligence IJCAI-89, volume 1, pages 768774,
Detroit, MI, USA, 20-25 August 1989. Morgan Kaufmann. URL http://www.
genetic-programming.com/jkpdf/ijcai1989.pdf.
GPBiB
J. R. Koza. Genetic programming: A paradigm for genetically breeding populations of computer programs to solve problems. Technical Report STAN-CS-901314, Dept. of Computer Science, Stanford University, June 1990. URL http:
//www.genetic-programming.com/jkpdf/tr1314.pdf.
GPBiB
J. R. Koza.
A genetic approach to econometric modeling.
Congress of the Econometric Society, Barcelona, Spain, 1990.
genetic-programming.com/jkpdf/wces1990.pdf.

In Sixth World
URL http://www.
GPBiB

J. R. Koza. Genetic Programming: On the Programming of Computers by Means of

Natural Selection. MIT Press, Cambridge, MA, USA, 1992. ISBN 0-262-11170-5.
GPBiB
J. R. Koza. Genetic Programming II: Automatic Discovery of Reusable Programs. MIT
Press, Cambridge Massachusetts, May 1994. ISBN 0-262-11189-6.
GPBiB
J. R. Koza. Two ways of discovering the size and shape of a computer program to solve a
problem. In L. Eshelman, editor, Genetic Algorithms: Proceedings of the Sixth International Conference (ICGA95), pages 287294, Pittsburgh, PA, USA, 15-19 July 1995.
Morgan Kaufmann. ISBN 1-55860-370-0. URL http://www.genetic-programming.
com/jkpdf/icga1995.pdf.
GPBiB
J. R. Koza, editor. Late Breaking Papers at the Genetic Programming 1996 Conference
Stanford University July 28-31, 1996, Stanford University, CA, USA, 2831 July 1996.
Stanford Bookstore. ISBN 0-18-201031-7. URL http://www.genetic-programming.
org/gp96latebreaking.html.
GPBiB
J. R. Koza, editor. Late Breaking Papers at the 1997 Genetic Programming Conference, Stanford University, CA, USA, 1316 July 1997. Stanford Bookstore. ISBN
0-18-206995-8. URL http://www.genetic-programming.org/gp97latebreaking.html.
GPBiB
J. R. Koza, editor. Late Breaking Papers at the 1998 Genetic Programming Conference,
University of Wisconsin, Madison, WI, USA, 22-25 July 1998. Omni Press.
GPBiB
J. R. Koza, S. H. Al-Sakran, and L. W. Jones. Automated re-invention of six patented
optical lens systems using genetic programming. In H.-G. Beyer, et al., editors, GECCO
2005: Proceedings of the 2005 conference on Genetic and evolutionary computation,
volume 2, pages 19531960, Washington DC, USA, 25-29 June 2005. ACM Press.
ISBN 1-59593-010-8. URL http://www.cs.bham.ac.uk/~wbl/biblio/gecco2005/docs/
p1953.pdf.
GPBiB
J. R. Koza and D. Andre. Classifying protein segments as transmembrane domains using
architecture-altering operations in genetic programming. In P. J. Angeline and K. E.
Kinnear, Jr., editors, Advances in Genetic Programming 2, chapter 8, pages 155176.
MIT Press, Cambridge, MA, USA, 1996. ISBN 0-262-01158-1. URL http://www.
genetic-programming.com/jkpdf/aigp2aatmjk1996.pdf.
GPBiB
J. R. Koza, D. Andre, F. H. Bennet, III, and M. Keane. Genetic Programming 3: Darwinian Invention and Problem Solving. Morgan Kaufman, April 1999. ISBN 1-55860543-6. URL http://www.genetic-programming.org/gpbook3toc.html.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=13)

1.5

80
1
100
0.5

60
Fitness

Generation 13
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

192

BIBLIOGRAPHY

J. R. Koza, D. Andre, F. H. Bennett, III, and M. A. Keane. Use of automatically

defined functions and architecture-altering operations in automated circuit synthesis
using genetic programming. In J. R. Koza, et al., editors, Genetic Programming 1996:
Proceedings of the First Annual Conference, pages 132149, Stanford University, CA,
USA, 2831 July 1996a. MIT Press. URL http://www.genetic-programming.com/
jkpdf/gp1996adfaa.pdf.
GPBiB
J. R. Koza, F. H. Bennett, III, D. Andre, and M. A. Keane. Four problems for
which a computer program evolved by genetic programming is competitive with
human performance. In Proceedings of the 1996 IEEE International Conference
on Evolutionary Computation, volume 1, pages 110. IEEE Press, 1996b. URL
http://www.genetic-programming.com/jkpdf/icec1996.pdf.
GPBiB
J. R. Koza, F. H. Bennett, III, D. Andre, and M. A. Keane. Automated WYWIWYG
design of both the topology and component values of electrical circuits using genetic
programming. In J. R. Koza, et al., editors, Genetic Programming 1996: Proceedings of the First Annual Conference, pages 123131, Stanford University, CA, USA,
2831 July 1996c. MIT Press. URL http://www.genetic-programming.com/jkpdf/
gp1996nielsen.pdf.
GPBiB
J. R. Koza, F. H. Bennett, III, D. Andre, and M. A. Keane. The design of analog circuits
by means of genetic programming. In P. Bentley, editor, Evolutionary Design by
Computers, chapter 16, pages 365385. Morgan Kaufmann, San Francisco, USA, 1999.
ISBN 1-55860-605-X. URL http://www.genetic-programming.com/jkpdf/edc1999.
pdf.
GPBiB
J. R. Koza, F. H. Bennett, III, J. L. Hutchings, S. L. Bade, M. A. Keane, and D. Andre. Evolving sorting networks using genetic programming and the rapidly reconfigurable xilinx 6216 field-programmable gate array. In Proceedings of the 31st Asilomar Conference on Signals, Systems, and Computers. IEEE Press, 1997. URL
http://www.genetic-programming.com/jkpdf/ieeeasilmoar1997.pdf.
GPBiB
J. R. Koza, F. H. Bennett, III, and O. Stiffelman. Genetic programming as a Darwinian
invention machine. In R. Poli, et al., editors, Genetic Programming, Proceedings of
EuroGP99, volume 1598 of LNCS, pages 93108, Goteborg, Sweden, 26-27 May 1999.
Springer-Verlag. ISBN 3-540-65899-8. URL http://www.genetic-programming.com/
jkpdf/eurogp1999.pdf.
GPBiB
J. R. Koza, L. W. Jones, M. A. Keane, and M. J. Streeter. Towards industrial strength
automated design of analog electrical circuits by means of genetic programming. In U.M. OReilly, et al., editors, Genetic Programming Theory and Practice II, chapter 8,
pages 120?? Springer, Ann Arbor, 13-15 May 2004. ISBN 0-387-23253-2. URL
http://www.genetic-programming.com/gptp2004.pdf. pages missing?
GPBiB
J. R. Koza, M. A. Keane, M. J. Streeter, W. Mydlowec, J. Yu, and G. Lanza. Genetic Programming IV: Routine Human-Competitive Machine Intelligence. Kluwer Academic
Publishers, 2003. ISBN 1-4020-7446-8. URL http://www.genetic-programming.org/
gpbook4toc.html.
GPBiB
J. R. Koza and R. Poli. Genetic programming. In E. K. Burke and G. Kendall,
editors, Search Methodologies: Introductory Tutorials in Optimization and Decision
Support Techniques, chapter 5. Springer, 2005. ISBN 0-387-23460-8. URL http://
www.springer.com/sgw/cda/frontpage/0,11855,4-10045-22-67933962-0,00.html.
GPBiB

BIBLIOGRAPHY

193

J. R. Koza, D. E. Goldberg, D. B. Fogel, and R. L. Riolo, editors. Genetic Programming 1996: Proceedings of the First Annual Conference, Stanford University, CA,
USA, 2831 July 1996. MIT Press. URL http://www.genetic-programming.org/
gp96proceedings.html.
GPBiB
J. R. Koza, K. Deb, M. Dorigo, D. B. Fogel, M. Garzon, H. Iba, and R. L. Riolo, editors.
Genetic Programming 1997: Proceedings of the Second Annual Conference, Stanford
University, CA, USA, 13-16 July 1997. Morgan Kaufmann. URL http://www.mkp.com/
books_catalog/1-55860-483-9.asp.
GPBiB
J. R. Koza, W. Banzhaf, K. Chellapilla, K. Deb, M. Dorigo, D. B. Fogel, M. H. Garzon,
D. E. Goldberg, H. Iba, and R. Riolo, editors. Genetic Programming 1998: Proceedings
of the Third Annual Conference, University of Wisconsin, Madison, WI, USA, 22-25
July 1998. Morgan Kaufmann. ISBN 1-55860-548-7.
GPBiB
D. H. Kraft, F. E. Petry, W. P. Buckles, and T. Sadasivan. The use of genetic programming
to build queries for information retrieval. In Proceedings of the 1994 IEEE World
Congress on Computational Intelligence, pages 468473, Orlando, Florida, USA, 2729 June 1994. IEEE Press.
GPBiB
T. Krantz, O. Lindberg, G. Thorburn, and P. Nordin. Programmatic compression of
natural video. In E. Cant
u-Paz, editor, Late Breaking Papers at the Genetic and Evolutionary Computation Conference (GECCO-2002), pages 301307, New York, NY,
July 2002. AAAI. URL http://thomas.krantz.com/paper.pdf.
GPBiB
N. Krasnogor. Self generating metaheuristics in bioinformatics: The proteins structure
comparison case. Genetic Programming and Evolvable Machines, 5(2):181201, June
2004. ISSN 1389-2576.
K. Krawiec. Evolutionary Feature Programming: Cooperative learning for knowledge
discovery and computer vision. Number 385. Wydawnictwo Politechniki Poznanskiej,
Poznan University of Technology, Poznan, Poland, 2004. URL http://idss.cs.put.
poznan.pl/~krawiec/pubs/hab/krawiec_hab.pdf.
GPBiB
W. B. Langdon. The evolution of size in variable length representations. In 1998 IEEE
International Conference on Evolutionary Computation, pages 633638, Anchorage,
Alaska, USA, 5-9 May 1998. IEEE Press. URL http://www.cs.bham.ac.uk/~wbl/ftp/
papers/WBL.wcci98_bloat.pdf.
GPBiB
W. B. Langdon. Size fair and homologous tree genetic programming crossovers. In
W. Banzhaf, et al., editors, Proceedings of the Genetic and Evolutionary Computation
Conference, volume 2, pages 10921097, Orlando, Florida, USA, 13-17 July 1999a.
Morgan Kaufmann. ISBN 1-55860-611-4. URL http://www.cs.ucl.ac.uk/staff/W.
Langdon/ftp/papers/WBL.gecco99.fairxo.ps.gz.
GPBiB
W. B. Langdon. Scaling of program tree fitness spaces. Evolutionary Computation, 7(4):
399428, Winter 1999b. ISSN 1063-6560. URL http://www.mitpressjournals.org/
doi/pdf/10.1162/evco.1999.7.4.399.
GPBiB
W. B. Langdon. Convergence rates for the distribution of program outputs. In W. B.
Langdon, et al., editors, GECCO 2002: Proceedings of the Genetic and Evolutionary Computation Conference, pages 812819, New York, 9-13 July 2002a. Morgan
Kaufmann Publishers. ISBN 1-55860-878-8. URL http://www.cs.ucl.ac.uk/staff/
W.Langdon/ftp/papers/wbl_gecco2002.pdf.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=15)

1.5

80
1
100
0.5

60
Fitness

Generation 15
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

194

BIBLIOGRAPHY

W. B. Langdon. How many good programs are there? How long are they? In K. A.
De Jong, et al., editors, Foundations of Genetic Algorithms VII, pages 183202,
Torremolinos, Spain, 4-6 September 2002b. Morgan Kaufmann. ISBN 0-12-2081552. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/wbl_foga2002.pdf.
Published 2003.
GPBiB
W. B. Langdon. Convergence of program fitness landscapes. In E. Cant
u-Paz, et al., editors, Genetic and Evolutionary Computation GECCO-2003, volume 2724 of LNCS,
pages 17021714, Chicago, 12-16 July 2003a. Springer-Verlag. ISBN 3-540-406034.
URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/wbl_gecco2003.
pdf.
GPBiB
W. B. Langdon. The distribution of reversible functions is Normal. In R. L. Riolo and
B. Worzel, editors, Genetic Programming Theory and Practise, chapter 11, pages 173
188. Kluwer, 2003b. ISBN 1-4020-7581-2. URL http://www.cs.ucl.ac.uk/staff/W.
Langdon/ftp/papers/wbl_reversible.pdf.
GPBiB
W. B. Langdon. Global distributed evolution of L-systems fractals. In M. Keijzer, et al.,
editors, Genetic Programming, Proceedings of EuroGP2004, volume 3003 of LNCS,
pages 349358, Coimbra, Portugal, 5-7 April 2004. Springer-Verlag. ISBN 3-540-213465. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/egp2004_pfeiffer.
pdf.
GPBiB
W. B. Langdon. Pfeiffer A distributed open-ended evolutionary system. In B. Edmonds,
et al., editors, AISB05: Proceedings of the Joint Symposium on Socially Inspired
Computing (METAS 2005), pages 713, University of Hertfordshire, Hatfield, UK, 1215 April 2005a. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/wbl_
metas2005.pdf. SSAISB 2005 Convention.
GPBiB
W. B. Langdon. The distribution of amorphous computer outputs. In S. Stepney and
S. Emmott, editors, The Grand Challenge in Non-Classical Computation: International Workshop, York, UK, 18-19 April 2005b. URL http://www.cs.ucl.ac.uk/
staff/W.Langdon/ftp/papers/grand_2005.pdf.
GPBiB
W. B. Langdon. Mapping non-conventional extensions of genetic programming. In C. S.
Calude, et al., editors, Unconventional Computing 2006, volume 4135 of LNCS, pages
166180, York, 4-8 September 2006. Springer-Verlag. ISBN 3-540-38593-2. URL http:
//www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/wbl_uc2002.pdf.
GPBiB
W. B. Langdon and W. Banzhaf. A SIMD interpreter for genetic programming on GPU
graphics cards. In EuroGP, LNCS, Naples, 26-28 March 2008. Springer. URL http://
www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/langdon_2008_eurogp.pdf. Forthcoming.
W. B. Langdon, S. J. Barrett, and B. F. Buxton. Predicting biochemical interactions
human P450 2D6 enzyme inhibition. In R. Sarker, et al., editors, Proceedings of
the 2003 Congress on Evolutionary Computation CEC2003, pages 807814, Canberra,
8-12 December 2003. IEEE Press. ISBN 0-7803-7804-0. URL http://www.cs.ucl.ac.
uk/staff/W.Langdon/ftp/papers/wbl_cec2003.pdf.
GPBiB
W. B. Langdon and B. F. Buxton. Genetic programming for mining DNA chip data
from cancer patients. Genetic Programming and Evolvable Machines, 5(3):251257,
September 2004. ISSN 1389-2576. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/
ftp/papers/wbl_dnachip.pdf.
GPBiB
W. B. Langdon and A. P. Harrison. GP on SPMD parallel graphics hardware for mega
bioinformatics data mining. 2008. In preparation.

BIBLIOGRAPHY

195

W. B. Langdon and J. P. Nordin. Seeding GP populations. In R. Poli, et al., editors,

Genetic Programming, Proceedings of EuroGP2000, volume 1802 of LNCS, pages 304
315, Edinburgh, 15-16 April 2000. Springer-Verlag. ISBN 3-540-67339-3. URL http://
www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/WBL_eurogp2000_seed.pdf. GPBiB
W. B. Langdon and R. Poli. Fitness causes bloat. In P. K. Chawdhry, et al., editors, Soft
Computing in Engineering Design and Manufacturing, pages 1322. Springer-Verlag
London, 23-27 June 1997. ISBN 3-540-76214-0. URL http://www.cs.bham.ac.uk/
~wbl/ftp/papers/WBL.bloat_wsc2.ps.gz.
GPBiB
W. B. Langdon and R. Poli. Why ants are hard. In J. R. Koza, et al., editors, Genetic
Programming 1998: Proceedings of the Third Annual Conference, pages 193201, University of Wisconsin, Madison, Wisconsin, USA, 22-25 July 1998a. Morgan Kaufmann.
ISBN 1-55860-548-7. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/
WBL.antspace_gp98.pdf.
GPBiB
W. B. Langdon and R. Poli. Better trained ants for genetic programming. Technical
Report CSRP-98-12, University of Birmingham, School of Computer Science, April
1998b. URL ftp://ftp.cs.bham.ac.uk/pub/tech-reports/1998/CSRP-98-12.ps.gz.
GPBiB
W. B. Langdon and R. Poli. The halting probability in von Neumann architectures.
In P. Collet, et al., editors, Proceedings of the 9th European Conference on Genetic
Programming, volume 3905 of Lecture Notes in Computer Science, pages 225237,
Budapest, Hungary, 10 - 12 April 2006. Springer. ISBN 3-540-33143-3. URL http:
//www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/wbl_egp2006.pdf.
GPBiB
W. B. Langdon and R. Poli. Foundations of Genetic Programming. Springer-Verlag,
2002. ISBN 3-540-42451-2. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/FOGP/.
GPBiB
W. B. Langdon and P. C. Treleaven. Scheduling maintenance of electrical power transmission networks using genetic programming. In K. Warwick, et al., editors, Artificial Intelligence Techniques in Power Systems, chapter 10, pages 220237. IEE,
1997. ISBN 0-85296-897-3. URL http://www.iee.org/Publish/Books/Power/Po022c.
cfm#10.Scheduling.
GPBiB
W. B. Langdon, R. Poli, P. Nordin, and T. Fogarty, editors. Late-Breaking Papers
of EuroGP-99, Goteborg, Sweden, 26-27 May 1999. URL ftp://ftp.cwi.nl/pub/
CWIreports/SEN/SEN-R9913.pdf.
GPBiB
W. B. Langdon, E. Cant
u-Paz, K. Mathias, R. Roy, D. Davis, R. Poli, K. Balakrishnan,
V. Honavar, G. Rudolph, J. Wegener, L. Bull, M. A. Potter, A. C. Schultz, J. F. Miller,
E. Burke, and N. Jonoska, editors. GECCO 2002: Proceedings of the Genetic and
Evolutionary Computation Conference, New York, 9-13 July 2002. Morgan Kaufmann
Publishers. ISBN 1-55860-878-8. URL http://www.isgec.org/GECCO-2002.
GPBiB
W. B. Langdon. A bibliography for genetic programming. In P. J. Angeline and K. E.
Kinnear, Jr., editors, Advances in Genetic Programming 2, chapter B, pages 507532.
MIT Press, Cambridge, MA, USA, 1996. ISBN 0-262-01158-1. URL http://www.cs.
ucl.ac.uk/staff/W.Langdon/ftp/papers/WBL.aigp2.appx.ps.gz.
GPBiB
W. B. Langdon. Genetic Programming and Data Structures. Kluwer, Boston, 1998. ISBN
0-7923-8135-1. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/gpdata.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=17)

1.5

80
1
100
0.5

60
Fitness

Generation 17
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

196

BIBLIOGRAPHY

W. B. Langdon. Size fair and homologous tree genetic programming crossovers. Genetic Programming and Evolvable Machines, 1(1/2):95119, April 2000. ISSN 13892576.
URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/WBL_fairxo.
pdf.
GPBiB
W. B. Langdon and W. Banzhaf. Repeated sequences in linear genetic programming
genomes. Complex Systems, 15(4):285306, 2005. ISSN 0891-2513. URL http://www.
cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/wbl_repeat_linear.pdf.
GPBiB
W. B. Langdon, S. M. Gustafson, and J. Koza. The Genetic Programming Bibliography,
1995-2008. URL http://www.cs.bham.ac.uk/~wbl/biblio/.
W. B. Langdon and P. Nordin. Evolving hand-eye coordination for a humanoid robot
with machine code genetic programming. In J. F. Miller, et al., editors, Genetic
Programming, Proceedings of EuroGP2001, volume 2038 of LNCS, pages 313324,
Lake Como, Italy, 18-20 April 2001. Springer-Verlag. ISBN 3-540-41899-7. URL http:
//www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/wbl_handeye.ps.gz.
GPBiB
W. B. Langdon and R. Poli. Evolutionary solo pong players. In D. Corne, et al., editors,
Proceedings of the 2005 IEEE Congress on Evolutionary Computation, volume 3, pages
26212628, Edinburgh, UK, 2-5 September 2005. IEEE Press. ISBN 0-7803-93635. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/pong_cec2005.pdf.
GPBiB
W. B. Langdon and R. Poli. On turing complete T7 and MISC F4 program fitness
landscapes. In D. V. Arnold, et al., editors, Theory of Evolutionary Algorithms, number 06061 in Dagstuhl Seminar Proceedings, Dagstuhl, Germany, 5-10 February 2006.
Internationales Begegnungs- und Forschungszentrum fuer Informatik (IBFI), Schloss
Dagstuhl, Germany. URL http://drops.dagstuhl.de/opus/volltexte/2006/595.
<http://drops.dagstuhl.de/opus/volltexte/2006/595> [date of citation: 2006-01-01].
GPBiB
W. B. Langdon and R. Poli. Mapping non-conventional extensions of genetic programming. Natural Computing, 7:2143, March 2008. Invited contribution to special issue
on Unconventional computing.
GPBiB
W. B. Langdon, T. Soule, R. Poli, and J. A. Foster. The evolution of size and shape.
In L. Spector, et al., editors, Advances in Genetic Programming 3, chapter 8, pages
163190. MIT Press, Cambridge, MA, USA, June 1999. ISBN 0-262-19423-6. URL
http://www.cs.bham.ac.uk/~wbl/aigp3/ch08.pdf.
GPBiB
P. Larra
naga. A review on estimation of distribution algorithms, chapter 3, pages 57100.
Kluwer Academic Publishers, 2002.
P. Larra
naga and J. A. Lozano. Estimation of Distribution Algorithms, A New Tool for
Evolutionary Computation. Kluwer Academic Publishers, 2002.
S. Lavington, N. Dewhurst, E. Wilkins, and A. Freitas. Interfacing knowledge discovery
algorithms to large database management systems. Information and Software Technology, 41(9):605617, 25 June 1999. URL http://www.sciencedirect.com/science/
article/B6V0B-3WN7DYN-8/1/cdabdda09c085c6a4536aa5e116366ee. special issue on
data mining.
GPBiB
K. S. Leung, K. H. Lee, and S. M. Cheang. Genetic parallel programming - evolving linear
machine codes on a multiple-ALU processor. In S. Yaacob, et al., editors, Proceedings
of International Conference on Artificial Intelligence in Engineering and Technology ICAIET 2002, pages 207213. Universiti Malaysia Sabah, June 2002. ISBN 983-218892-X.
GPBiB

BIBLIOGRAPHY

197

T. L. Lew, A. B. Spencer, F. Scarpa, K. Worden, A. Rutherford, and F. Hemez. Identification of response surface models using genetic programming. Mechanical Systems
and Signal Processing, 20(8):18191831, November 2006.
GPBiB
D. R. Lewin, S. Lachman-Shalem, and B. Grosman. The role of process system engineering
(PSE) in integrated circuit (IC) manufacturing. Control Engineering Practice, 15(7):
793802, July 2006. Special Issue on Award Winning Applications, 2005 IFAC World
Congress.
GPBiB
J. Li and E. P. K. Tsang. Investment decision making using FGP: A case study. In P. J.
Angeline, et al., editors, Proceedings of the Congress on Evolutionary Computation,
volume 2, pages 12531259, Mayflower Hotel, Washington D.C., USA, 6-9 July 1999.
IEEE Press. ISBN 0-7803-5536-9 (softbound). URL http://www.cs.bham.ac.uk/~jxl/
cercialink/web/publication/CEC99.pdf.
GPBiB
L. Li, W. Jiang, X. Li, K. L. Moser, Z. Guo, L. Du, Q. Wang, E. J. Topol, Q. Wang,
and S. Rao. A robust hybrid between genetic algorithm and support vector machine
for extracting an optimal feature gene subset. Genomics, 85(1):1623, January 2005.
GPBiB
R. Linden and A. Bhaya. Evolving fuzzy rules to model gene expression. Biosystems, 88
(1-2):7691, March 2007.
GPBiB
A. Lindenmayer. Mathematic models for cellular interaction in development, parts I and
II. Journal of Theoretical Biology, 18:280299 and 300315, 1968.
H. Lipson. How to draw a straight line using a GP: Benchmarking evolutionary design
against 19th century kinematic synthesis. In M. Keijzer, editor, Late Breaking Papers
at the 2004 Genetic and Evolutionary Computation Conference, Seattle, Washington, USA, 26 July 2004. URL http://www.cs.bham.ac.uk/~wbl/biblio/gecco2004/
LBP063.pdf.
GPBiB
J. Lohn, G. Hornby, and D. Linden. Evolutionary antenna design for a NASA spacecraft.
In U.-M. OReilly, et al., editors, Genetic Programming Theory and Practice II, chapter 18, pages 301315. Springer, Ann Arbor, 13-15 May 2004. ISBN 0-387-23253-2.
GPBiB
J. Lohn, A. Stoica, and D. Keymeulen, editors. The Second NASA/DoD Workshop on
Evolvable Hardware, Palo Alto, California, 13-15 July 2000. IEEE Computer Society.
ISBN 0-7695-0762-X.
M. A. Lones. Enzyme Genetic Programming: Modelling Biological Evolvability in Genetic
Programming. PhD thesis, The University of York, Heslington, York, YO10 5DD, UK,
September 2003. URL http://www-users.york.ac.uk/~mal503/common/thesis/main.
html.
GPBiB
M. Looks. Scalable estimation-of-distribution program evolution. In H. Lipson, editor,
GECCO, pages 539546. ACM, 2007. ISBN 978-1-59593-697-4.
M. Looks, B. Goertzel, and C. Pennachin. Learning computer programs with the bayesian
optimization algorithm. In H.-G. Beyer, et al., editors, GECCO 2005: Proceedings of
the 2005 conference on Genetic and evolutionary computation, volume 1, pages 747
748, Washington DC, USA, 25-29 June 2005. ACM Press. ISBN 1-59593-010-8. URL
http://www.cs.bham.ac.uk/~wbl/biblio/gecco2005/docs/p747.pdf.
GPBiB
J. Louchet. Using an individual evolution strategy for stereovision. Genetic Programming
and Evolvable Machines, 2(2):101109, June 2001. ISSN 1389-2576.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=19)

1.5

80
1
100
0.5

60
Fitness

Generation 19
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

198

BIBLIOGRAPHY

J. Louchet, M. Guyon, M.-J. Lesot, and A. Boumaza. Dynamic flies: a new pattern
recognition tool applied to stereo sequence processing. Pattern Recognition Letters, 23
(1-3):335345, January 2002.
GPBiB
J. Loviscach and J. Meyer-Spradow. Genetic programming of vertex shaders. In
M. Chover, et al., editors, Proceedings of EuroMedia 2003, pages 2931, 2003. GPBiB
S. Luke. Evolving soccerbots: A retrospective. In Proceedings of the 12th Annual
Conference of the Japanese Society for Artificial Intelligence, 1998. URL http:
//www.cs.gmu.edu/~sean/papers/robocupShort.pdf.
GPBiB
S. Luke. Two fast tree-creation algorithms for genetic programming. IEEE Transactions on Evolutionary Computation, 4(3):274283, September 2000. URL http:
//ieeexplore.ieee.org/iel5/4235/18897/00873237.pdf.
GPBiB
S. Luke and L. Panait. Lexicographic parsimony pressure. In W. B. Langdon, et al., editors, GECCO 2002: Proceedings of the Genetic and Evolutionary Computation Conference, pages 829836, New York, 9-13 July 2002. Morgan Kaufmann Publishers. ISBN
1-55860-878-8. URL http://cs.gmu.edu/~sean/papers/lexicographic.pdf. GPBiB
S. Luke, L. Panait, G. Balan, S. Paus, Z. Skolicki, E. Popovici, J. Harrison, J. Bassett,
R. Hubley, and A. Chircop. ECJ: A Java-based Evolutionary Computation Research
System , 2000-2007. URL http://www.cs.gmu.edu/~eclab/projects/ecj/.
S. Luke and L. Spector. A comparison of crossover and mutation in genetic programming. In J. R. Koza, et al., editors, Genetic Programming 1997: Proceedings of the
Second Annual Conference, pages 240248, Stanford University, CA, USA, 13-16 July
1997. Morgan Kaufmann. URL http://www.cs.gmu.edu/~sean/papers/comparison/
comparison.pdf.
GPBiB
E. Lukschandl, H. Borgvall, L. Nohle, M. Nordahl, and P. Nordin. Distributed java
bytecode genetic programming. In R. Poli, et al., editors, Genetic Programming, Proceedings of EuroGP2000, volume 1802 of LNCS, pages 316325, Edinburgh, 15-16
April 2000. Springer-Verlag. ISBN 3-540-67339-3. URL http://www.springerlink.
com/openurl.asp?genre=article&issn=0302-9743&volume=1802&spage=316. GPBiB
E. Lutton, J. Levy-Vehel, G. Cretin, P. Glevarec, and C. Roll. Mixed IFS: Resolution of
the inverse problem using genetic programming. Complex Systems, 9:375398, 1995a.
GPBiB
E. Lutton, J. Levy-Vehel, G. Cretin, P. Glevarec, and C. Roll. Mixed IFS: Resolution
of the inverse problem using genetic programming. Research Report No 2631, Inria,
1995b. URL http://citeseer.ist.psu.edu/cretin95mixed.html.
GPBiB
R. M. MacCallum. Introducing a perl genetic programming system: and can metaevolution solve the bloat problem? In C. Ryan, et al., editors, Genetic Programming,
Proceedings of EuroGP2003, volume 2610 of LNCS, pages 364373, Essex, 14-16 April
2003. Springer-Verlag. ISBN 3-540-00971-X. URL http://www.sbc.su.se/~maccallr/
publications/perlgp_eurogp2003.pdf.
GPBiB
P. Machado and J. Romero, editors. The Art of Artificial Evolution. Springer, 2008.
A. J. Marek, W. D. Smart, and M. C. Martin. Learning visual feature detectors for
obstacle avoidance using genetic programming. In E. Cant
u-Paz, editor, Late Breaking
Papers at the Genetic and Evolutionary Computation Conference (GECCO-2002),
pages 330336, New York, NY, July 2002. AAAI. URL http://www.martincmartin.
com/papers/LearingVisualFeatureDetectorsForObstAvoidGP_GECCO2002Marek.pdf.
GPBiB

BIBLIOGRAPHY

199

P. Marenbach. Using prior knowledge and obtaining process insight in data based modelling of bioprocesses. System Analysis Modelling Simulation, 31:3959, 1998. GPBiB
J. P. Marney, D. Miller, C. Fyfe, and H. F. E. Tarbert. Risk adjusted returns to technical
trading rules: a genetic programming approach. In 7th International Conference of
Society of Computational Economics, Yale, 28-29 June 2001.
GPBiB
M. C. Martin. Evolving visual sonar: Depth from monocular images. Pattern Recognition
Letters, 27(11):11741180, August 2006. URL http://martincmartin.com/papers/
EvolvingVisualSonarPatternRecognitionLetters2006.pdf. Evolutionary Computer
Vision and Image Understanding.
GPBiB
P. Martin. A hardware implementation of a genetic programming system using FPGAs
and Handel-C. Genetic Programming and Evolvable Machines, 2(4):317343, December 2001. ISSN 1389-2576. URL http://www.naiadhome.com/gpem-d.pdf.
GPBiB
P. Martin. A pipelined hardware implementation of genetic programming using FPGAs
and Handel-C. In J. A. Foster, et al., editors, Genetic Programming, Proceedings of the
5th European Conference, EuroGP 2002, volume 2278 of LNCS, pages 112, Kinsale,
Ireland, 3-5 April 2002. Springer-Verlag. ISBN 3-540-43378-3.
GPBiB
P. Martin and R. Poli. Crossover operators for A hardware implementation of GP using
FPGAs and Handel-C. In W. B. Langdon, et al., editors, GECCO 2002: Proceedings
of the Genetic and Evolutionary Computation Conference, pages 845852, New York,
9-13 July 2002. Morgan Kaufmann Publishers. ISBN 1-55860-878-8. URL http://
www.cs.bham.ac.uk/~wbl/biblio/gecco2002/gp284.ps.
GPBiB
S. Martinez-Jaramillo and E. P. K. Tsang. An heterogeneous, endogenous and coevolutionary GP-based financial market. IEEE Transactions on Evolutionary Computation, 2007. accepted for publication.
P. Massey, J. A. Clark, and S. Stepney. Evolution of a human-competitive quantum
fourier transform algorithm using genetic programming. In H.-G. Beyer, et al., editors, GECCO 2005: Proceedings of the 2005 conference on Genetic and evolutionary
computation, volume 2, pages 16571663, Washington DC, USA, 25-29 June 2005.
ACM Press. ISBN 1-59593-010-8. URL http://www.cs.bham.ac.uk/~wbl/biblio/
gecco2005/docs/p1657.pdf.
GPBiB
S. R. Maxwell, III. Why might some problems be difficult for genetic programming to find
solutions? In J. R. Koza, editor, Late Breaking Papers at the Genetic Programming
1996 Conference Stanford University July 28-31, 1996, pages 125128, Stanford University, CA, USA, 2831 July 1996. Stanford Bookstore. ISBN 0-18-201031-7. GPBiB
S. R. Maxwell, III. Experiments with a coroutine model for genetic programming. In
Proceedings of the 1994 IEEE World Congress on Computational Intelligence, volume 1, pages 413417a, Orlando, Florida, USA, 27-29 June 1994. IEEE Press. ISBN
0-7803-1899-4. URL http://ieeexplore.ieee.org/iel2/1125/8059/00349915.pdf?
isNumber=8059.
GPBiB
J. McCormack. New challenges for evolutionary music and art. SIGEvolution, 1(1):511,
April 2006. URL http://www.sigevolution.org/2006/01/issue.pdf.
GPBiB
A. C. McGovern, D. Broadhurst, J. Taylor, N. Kaderbhai, M. K. Winson, D. A. Small,
J. J. Rowland, D. B. Kell, and R. Goodacre. Monitoring of complex industrial bioprocesses for metabolite concentrations using modern spectroscopies and machine learning:
Application to gibberellic acid production. Biotechnology and Bioengineering, 78(5):
527538, 5 June 2002. URL http://dbkgroup.org/Papers/biotechnol_bioeng_78_
(527).pdf.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=21)

1.5

80
1
100
0.5

60
Fitness

Generation 21
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

200

BIBLIOGRAPHY

B. McKay, M. Willis, D. Searson, and G. Montague. Nonlinear continuum regression:

an evolutionary approach. Transactions of the Institute of Measurement and Control,
22(2):125140, 2000. URL http://www.ingentaconnect.com/content/arn/tm/2000/
00000022/00000002/art00007.
GPBiB
B. McKay, M. J. Willis, and G. W. Barton.
Using a tree structured genetic algorithm to perform symbolic regression.
In A. M. S. Zalzala,
editor, First International Conference on Genetic Algorithms in Engineering Systems:
Innovations and Applications, GALESIA, volume 414, pages
487492, Sheffield, UK, 12-14 September 1995. IEE.
ISBN 0-85296-650-4.
URL http://scitation.aip.org/getpdf/servlet/GetPDFServlet?filetype=pdf&id=
IEECPS0019950CP414000487000001&idtype=cvips&prog=normal.
GPBiB
N. F. McPhee and N. J. Hopper. Analysis of genetic diversity through population history. In W. Banzhaf, et al., editors, Proceedings of the Genetic and Evolutionary
Computation Conference, volume 2, pages 11121120, Orlando, Florida, USA, 13-17
July 1999. Morgan Kaufmann. ISBN 1-55860-611-4. URL http://www.cs.bham.ac.
uk/~wbl/biblio/gecco1999/GP-421.pdf.
GPBiB
N. F. McPhee, N. J. Hopper, and M. L. Reierson. Sutherland: An extensible objectoriented software framework for evolutionary computation. In J. R. Koza, et al., editors, Genetic Programming 1998: Proceedings of the Third Annual Conference, page
241, University of Wisconsin, Madison, Wisconsin, USA, 22-25 July 1998. Morgan
Kaufmann. ISBN 1-55860-548-7. URL http://www.mrs.umn.edu/~mcphee/Research/
Sutherland/sutherland_gp98_announcement.ps.gz.
GPBiB
N. F. McPhee, A. Jarvis, and E. F. Crane. On the strength of size limits in linear genetic
programming. In K. Deb, et al., editors, Genetic and Evolutionary Computation
GECCO-2004, Part II, volume 3103 of Lecture Notes in Computer Science, pages
593604, Seattle, WA, USA, 26-30 June 2004. Springer-Verlag. ISBN 3-540-223436. URL http://link.springer.de/link/service/series/0558/bibs/3103/31030593.
htm.
GPBiB
N. F. McPhee and J. D. Miller. Accurate replication in genetic programming. In
L. Eshelman, editor, Genetic Algorithms: Proceedings of the Sixth International Conference (ICGA95), pages 303309, Pittsburgh, PA, USA, 15-19 July 1995. Morgan
Kaufmann. ISBN 1-55860-370-0. URL http://www.mrs.umn.edu/~mcphee/Research/
Accurate_replication.ps.
GPBiB
N. F. McPhee, B. Ohs, and T. Hutchison. Semantic building blocks in genetic programming. In EurogGP 2008: Proceedings of the 2008 European Conference on Genetic
Programming, 2008.
N. F. McPhee and R. Poli. A schema theory analysis of the evolution of size in genetic programming with linear representations. In J. F. Miller, et al., editors, Genetic Programming, Proceedings of EuroGP2001, volume 2038 of LNCS, pages 108125, Lake Como,
Italy, 18-20 April 2001. Springer-Verlag. ISBN 3-540-41899-7. URL http://cswww.
essex.ac.uk/staff/poli/papers/McPhee-EUROGP2001-ST-Linear-Bloat.pdf. GPBiB
N. F. McPhee and R. Poli. Using schema theory to explore interactions of multiple
operators. In W. B. Langdon, et al., editors, GECCO 2002: Proceedings of the Genetic
and Evolutionary Computation Conference, pages 853860, New York, 9-13 July 2002.
Morgan Kaufmann Publishers. ISBN 1-55860-878-8. URL http://www.cs.bham.ac.
uk/~wbl/biblio/gecco2002/GP139.pdf.
GPBiB

BIBLIOGRAPHY

201

R. R. F. Mendes, F. de B. Voznika, J. C. Nievola, and A. A. Freitas. Discovering fuzzy

classification rules with genetic programming and co-evolution. In L. Spector, et al., editors, Proceedings of the Genetic and Evolutionary Computation Conference (GECCO2001), page 183, San Francisco, California, USA, 7-11 July 2001. Morgan Kaufmann.
ISBN 1-55860-774-9. URL http://www.cs.bham.ac.uk/~wbl/biblio/gecco2001/d02.
pdf.
GPBiB
P. K. Mercure, G. F. Smits, and A. Kordon.
Empirical emulators for first
principle models.
In AIChE Fall Annual Meeting, Reno Hilton, 6 November 2001. URL http://www.aiche.org/conferences/techprogram/paperdetail.asp?
PaperID=2373&DSN=annual01.
GPBiB
J. Meyer-Spradow and J. Loviscach. Evolutionary design of BRDFs. In M. Chover, et al.,
editors, Eurographics 2003 Short Paper Proceedings, pages 301306, 2003. URL http:
//viscg.uni-muenster.de/publications/2003/ML03/evolutionary_web.pdf. GPBiB
J. Miller, M. Tomassini, P. L. Lanzi, C. Ryan, A. G. B. Tettamanzi, and W. B. Langdon,
editors. Genetic Programming, Proceedings of EuroGP2001, volume 2038 of LNCS,
Lake Como, Italy, 18-20 April 2001. Springer-Verlag. ISBN 3-540-41899-7. URL http:
//link.springer.de/link/service/series/0558/tocs/t2038.htm.
GPBiB
J. F. Miller. An empirical study of the efficiency of learning boolean functions using a
cartesian genetic programming approach. In W. Banzhaf, et al., editors, Proceedings of
the Genetic and Evolutionary Computation Conference, volume 2, pages 11351142,
Orlando, Florida, USA, 13-17 July 1999. Morgan Kaufmann. ISBN 1-55860-611-4.
URL http://citeseer.ist.psu.edu/153431.html.
GPBiB
J. F. Miller and S. L. Smith. Redundancy and computational efficiency in cartesian genetic
programming. IEEE Transactions on Evolutionary Computation, 10(2):167174, April
2006.
GPBiB
J. F. Miller, A. Thompson, P. Thomson, and T. C. Fogarty, editors. Proceedings of
the Third International Conference on Evolvable Systems, ICES 2000, volume 1801
of LNCS, Edinburgh, Scotland, UK, 17-19 April 2000. Springer-Verlag. ISBN 3-54067338-5.
B. Mitavskiy and J. Rowe. Some results about the markov chains associated to GPs and
to general EAs. Theoretical Computer Science, 361(1):72110, 28 August 2006. GPBiB
D. J. Montana. Strongly typed genetic programming. Evolutionary Computation, 3(2):
199230, 1995. URL http://vishnu.bbn.com/papers/stgp.pdf.
GPBiB
G. E. Moore.
Cramming more components onto integrated circuits.
Electronics, 38(8):114117, 1965.
URL ftp://download.intel.com/museum/Moores_Law/
Articles-Press_Releases/Gordon_Moore_1965_Article.pdf.
J. H. Moore, J. S. Parker, N. J. Olsen, and T. M. Aune. Symbolic discriminant analysis
of microarray data in automimmune disease. Genetic Epidemiology, 23:5769, 2002.
GPBiB
A. A. Motsinger, S. L. Lee, G. Mellick, and M. D. Ritchie. GPNN: Power studies and
applications of a neural network method for detecting gene-gene interactions in studies
of human disease. BMC bioinformatics [electronic resource], 7(1):3939, January 25
2006. ISSN 1471-2105. URL http://www.biomedcentral.com/1471-2105/7/39. GPBiB
H. M
uhlenbein and T. Mahnig. Convergence theory and application of the factorized
distribution algorithm. Journal of Computing and Information Technology, 7(1):19
32, 1999a.
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=23)

1.5

80
1
100
0.5

60
Fitness

Generation 23
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

202

BIBLIOGRAPHY

H. M
uhlenbein and T. Mahnig. FDA a scalable evolutionary algorithm for the optimization of additively decomposed functions. Evolutionary Computation, 7(4):353376,
1999b.
C. J. Neely. Risk-adjusted, ex ante, optimal technical trading rules in equity markets.
International Review of Economics and Finance, 12(1):6987, Spring 2003. URL http:
//research.stlouisfed.org/wp/1999/1999-015.pdf.
GPBiB
C. J. Neely and P. A. Weller. Technical trading rules in the european monetary system.
Journal of International Money and Finance, 18(3):429458, 1999. URL http://
research.stlouisfed.org/wp/1997/97-015.pdf.
GPBiB
C. J. Neely and P. A. Weller. Technical analysis and central bank intervention. Journal
of International Money and Finance, 20(7):949970, December 2001. URL http:
//research.stlouisfed.org/wp/1997/97-002.pdf.
GPBiB
C. J. Neely, P. A. Weller, and R. Dittmar.
Is technical analysis in the foreign exchange market profitable? A genetic programming approach. The Journal of Financial and Quantitative Analysis, 32(4):405426, December 1997. ISSN
00221090. URL http://links.jstor.org/sici?sici=0022-1090%28199712%2932%3A4%
3C405%3AITAITF%3E2.0.CO%3B2-T.
GPBiB
C. J. Neely, P. A. Weller, and J. M. Ulrich. The adaptive markets hypothesis: evidence
from the foreign exchange market. Working Paper 2006-046B, Federal Reserve Bank of
St. Louis, Research Division, P.O. Box 442, St. Louis, MO 63166, USA, August 2006.
URL http://research.stlouisfed.org/wp/2006/2006-046.pdf. Revised March 2007.
GPBiB
O. Nicolotti, V. J. Gillet, P. J. Fleming, and D. V. S. Green. Multiobjective optimization
in quantitative structure-activity relationships: Deriving accurate and interpretable
QSARs. Journal of Medicinal Chemistry, 45(23):50695080, November 7 2002. ISSN
0022-2623. URL http://pubs3.acs.org/acs/journals/doilookup?in_doi=10.1021/
jm020919o.
GPBiB
N. Nikolaev and H. Iba. Adaptive Learning of Polynomial Networks Genetic Programming, Backpropagation and Bayesian Methods. Number 4 in Genetic and Evolutionary
Computation. Springer, 2006. ISBN 0-387-31239-0. June.
GPBiB
N. Y. Nikolaev and H. Iba. Genetic programming of polynomial models for financial
forecasting. In S.-H. Chen, editor, Genetic Algorithms and Genetic Programming in
Computational Finance, chapter 5, pages 103123. Kluwer Academic Press, 2002. ISBN
0-7923-7601-3.
GPBiB
A. E. Nix and M. D. Vose. Modeling genetic algorithms with Markov chains. Annals of
Mathematics and Artificial Intelligence, 5:7988, 1992.
P. Nordin. A compiling genetic programming system that directly manipulates the machine code. In K. E. Kinnear, Jr., editor, Advances in Genetic Programming, chapter 14, pages 311331. MIT Press, 1994. URL http://cognet.mit.edu/library/books/
view?isbn=0262111888.
GPBiB
P. Nordin. Evolutionary Program Induction of Binary Machine Code and its Applications.
PhD thesis, der Universitat Dortmund am Fachereich Informatik, 1997.
GPBiB
P. Nordin and W. Banzhaf. Programmatic compression of images and sound. In J. R.
Koza, et al., editors, Genetic Programming 1996: Proceedings of the First Annual
Conference, pages 345350, Stanford University, CA, USA, 2831 July 1996. MIT
Press. URL http://www.cs.mun.ca/~banzhaf/papers/gp96.pdf.
GPBiB

BIBLIOGRAPHY

203

P. Nordin, W. Banzhaf, and F. D. Francone. Efficient evolution of machine code for

CISC architectures using instruction blocks and homologous crossover. In L. Spector,
et al., editors, Advances in Genetic Programming 3, chapter 12, pages 275299. MIT
Press, Cambridge, MA, USA, June 1999. ISBN 0-262-19423-6. URL http://www.
aimlearning.com/aigp31.pdf.
GPBiB
P. Nordin and W. Johanna. Humanoider: Sjavlarande robotar och artificiell intelligens.
Liber, 2003.
GPBiB
nVidia. NVIDIA CUDA Compute Unified Device Architecture, programming guide. Technical Report version 0.8, NVIDIA, 12 Feb 2007.
H. Oakley. Two scientific applications of genetic programming: Stack filters and nonlinear equation fitting to chaotic data. In K. E. Kinnear, Jr., editor, Advances in
Genetic Programming, chapter 17, pages 369389. MIT Press, 1994. URL http://
cognet.mit.edu/library/books/view?isbn=0262111888.
GPBiB
J. R. Olsson. Inductive functional programming using incremental program transformation and Execution of logic programs by iterative-deepening A* SLD-tree search. Dr
scient thesis, University of Oslo, Norway, 1994.
J. R. Olsson. How to invent functions. In R. Poli, et al., editors, Genetic Programming,
Proceedings of EuroGP99, volume 1598 of LNCS, pages 232243, Goteborg, Sweden,
26-27 May 1999. Springer-Verlag. ISBN 3-540-65899-8. URL http://www.ia-stud.
hiof.no/~rolando/abstrart1.ps.
GPBiB
R.

R. Olsson.
Inductive functional programming using incremental
program transformation.
Artificial Intelligence,
74(1):5581,
March
1995.
URL
http://www.sciencedirect.com/science?_ob=MImg&_imagekey=
B6TYF-4002FJH-9-1&_cdi=5617&_orig=browse&_coverDate=03%2F31%2F1995&_
sk=999259998&wchp=dGLbVlb-lSzBV&_acct=C000010182&_version=1&_userid=
125795&md5=ba5db57b3fa83d990440da8dfd8afcd7&ie=f.pdf.
GPBiB

M. Oltean. Evolving evolutionary algorithms using linear genetic programming. Evolutionary Computation, 13(3):387410, Fall 2005. ISSN 1063-6560.
GPBiB
M. Oltean and D. Dumitrescu. Evolving TSP heuristics using multi expression programming. In M. Bubak, et al., editors, Computational Science - ICCS 2004: 4th International Conference, Part II, volume 3037 of Lecture Notes in Computer Science,
pages 670673, Krakow, Poland, 6-9 June 2004. Springer-Verlag. ISBN 3-540-221158.
URL http://springerlink.metapress.com/openurl.asp?genre=article&issn=
0302-9743&volume=3037&spage=670.
GPBiB
R. Ondas, M. Pelikan, and K. Sastry. Genetic programming, probabilistic incremental program evolution, and scalability. In J. Knowles, editor, WSC10: 10th Online
World Conference on Soft Computing in Industrial Applications, pages 363372, On
the World Wide Web, 19 September - 7 October 2005. ISBN 3-540-29123-7. URL
http://isxp1010c.sims.cranfield.ac.uk/Papers/paper122.pdf.
GPBiB
M. ONeill and C. Ryan. Grammatical Evolution: Evolutionary Automatic Programming
in a Arbitrary Language, volume 4 of Genetic programming. Kluwer Academic Publishers, 2003. ISBN 1-4020-7444-1. URL http://www.wkap.nl/prod/b/1-4020-7444-1.
GPBiB
M. ONeill, C. Ryan, M. Keijzer, and M. Cattolico. Crossover in grammatical evolution.
Genetic Programming and Evolvable Machines, 4(1):6793, March 2003. ISSN 13892576.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=25)

1.5

80
1
100
0.5

60
Fitness

Generation 25
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

204

BIBLIOGRAPHY

S. Openshaw and I. Turton. Building new spatial interaction models using genetic
programming. In T. C. Fogarty, editor, Evolutionary Computing, Lecture Notes
in Computer Science, Leeds, UK, 11-13 April 1994. Springer-Verlag. URL http:
//www.geog.leeds.ac.uk/papers/94-1/94-1.pdf.
GPBiB
U.-M. OReilly. An Analysis of Genetic Programming. PhD thesis, Carleton University, Ottawa-Carleton Institute for Computer Science, Ottawa, Ontario, Canada,
22 September 1995. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/
oreilly/abstract.ps.gz.
GPBiB
U.-M. OReilly. Investigating the generality of automatically defined functions. In J. R.
Koza, et al., editors, Genetic Programming 1996: Proceedings of the First Annual
Conference, pages 351356, Stanford University, CA, USA, 2831 July 1996. MIT
Press. URL http://citeseer.ist.psu.edu/24128.html.
GPBiB
U.-M. OReilly and M. Hemberg. Integrating generative growth and evolutionary computation for form exploration. Genetic Programming and Evolvable Machines, 8(2):
163186, June 2007. ISSN 1389-2576. Special issue on developmental systems. GPBiB
U.-M. OReilly and F. Oppacher. Program search with a hierarchical variable length
representation: Genetic programming, simulated annealing and hill climbing. In
Y. Davidor, et al., editors, Parallel Problem Solving from Nature PPSN III, number
866 in Lecture Notes in Computer Science, pages 397406, Jerusalem, 9-14 October
1994a. Springer-Verlag. ISBN 3-540-58484-6. URL http://www.cs.ucl.ac.uk/staff/
W.Langdon/ftp/papers/ppsn-94.ps.gz.
GPBiB
U.-M. OReilly and F. Oppacher. The troubling aspects of a building block hypothesis for genetic programming.
In L. D. Whitley and M. D. Vose, editors, Foundations of Genetic Algorithms 3, pages 7388, Estes Park, Colorado, USA, 31 July2 August 1994b. Morgan Kaufmann. ISBN 1-55860-356-5.
URL http://citeseer.ist.psu.edu/cache/papers/cs/163/http:zSzzSzwww.ai.mit.
eduzSzpeoplezSzunamayzSzpaperszSzfoga.pdf/oreilly92troubling.pdf. Published
1995.
GPBiB
U.-M. OReilly, T. Yu, R. L. Riolo, and B. Worzel, editors. Genetic Programming Theory
and Practice II, volume 8 of Genetic Programming, Ann Arbor, MI, USA, 13-15 May
2004. Springer. ISBN 0-387-23253-2. URL http://www.springeronline.com/sgw/cda/
frontpage/0,11855,5-40356-22-34954683-0,00.html.
GPBiB
M. Oussaid`
ene, B. Chopard, O. V. Pictet, and M. Tomassini. Parallel genetic programming and its application to trading model induction. Parallel Computing, 23(8):1183
1198, August 1997. URL http://citeseer.ist.psu.edu/cache/papers/cs/166/http:
zSzzSzlslwww.epfl.chzSz~marcozSzparcomp.pdf/oussaidene97parallel.pdf. GPBiB
J. D. Owens, D. Luebke, N. Govindaraju, M. Harris, J. Kruger, A. E. Lefohn, and T. J.
Purcell. A survey of general-purpose computation on graphics hardware. Computer
Graphics Forum, 26(1):80113, March 2007.
L. Panait and S. Luke. Alternative bloat control methods. In K. Deb, et al., editors,
Genetic and Evolutionary Computation GECCO-2004, Part II, volume 3103 of Lecture Notes in Computer Science, pages 630641, Seattle, WA, USA, 26-30 June 2004.
Springer-Verlag. ISBN 3-540-22343-6. URL http://cs.gmu.edu/~lpanait/papers/
panait04alternative.pdf.
GPBiB
J. Parent and A. Nowe. Evolving compression preprocessors with genetic programming.
In W. B. Langdon, et al., editors, GECCO 2002: Proceedings of the Genetic and
Evolutionary Computation Conference, pages 861867, New York, 9-13 July 2002.

BIBLIOGRAPHY

205

Morgan Kaufmann Publishers. ISBN 1-55860-878-8. URL http://www.cs.bham.ac.

uk/~wbl/biblio/gecco2002/gp256.ps.
GPBiB
D. Parrott, X. Li, and V. Ciesielski. Multi-objective techniques in genetic programming
for evolving classifiers. In D. Corne, et al., editors, Proceedings of the 2005 IEEE
Congress on Evolutionary Computation, volume 2, pages 11411148, Edinburgh, UK,
2-5 September 2005. IEEE Press. ISBN 0-7803-9363-5. URL http://goanna.cs.rmit.
edu.au/~xiaodong/publications/183.pdf.
GPBiB
M. Pelikan, D. E. Goldberg, and E. Cant
u-Paz. BOA: The Bayesian optimization algorithm. In W. Banzhaf, et al., editors, Proc. of the Genetic and Evolutionary Computation Conference GECCO-99, volume I, pages 525532, Orlando, FL, 13-17 1999.
Morgan Kaufmann Publishers, San Fransisco, CA. ISBN 1-55860-611-4.
T. Perkis. Stack-based genetic programming. In Proceedings of the 1994 IEEE World
Congress on Computational Intelligence, volume 1, pages 148153, Orlando, Florida,
USA, 27-29 June 1994. IEEE Press. URL http://citeseer.ist.psu.edu/432690.html.
GPBiB
W. Piaseczny, H. Suzuki, and H. Sawai. Chemical genetic programming coevolution
between genotypic strings and phenotypic trees. In K. Deb, et al., editors, Genetic and
Evolutionary Computation GECCO-2004, Part II, volume 3103 of Lecture Notes
in Computer Science, pages 715716, Seattle, WA, USA, 26-30 June 2004. SpringerVerlag. ISBN 3-540-22343-6. URL http://link.springer.de/link/service/series/
0558/bibs/3103/31030715.htm.
GPBiB
N. Pillay. Evolving solutions to ASCII graphics programming problems in intelligent
programming tutors. In R. Akerkar, editor, International Conference on Applied Artificial Intelligence (ICAAI2003), pages 236243, Fort Panhala, Kolhapur, India, 15-16
December 2003. TMRF. ISBN 81-901918-0-2.
GPBiB
H. Pohlheim. Visualization of evolutionary algorithms - set of standard techniques and
multidimensional visualization. In W. Banzhaf, et al., editors, Proceedings of the
Genetic and Evolutionary Computation Conference, volume 1, pages 533540, Orlando, Florida, USA, 13-17 July 1999. Morgan Kaufmann. ISBN 1-55860-611-4. URL
http://www.cs.bham.ac.uk/~wbl/biblio/gecco1999/GA-820.pdf.
R. Poli. Discovery of symbolic, neuro-symbolic and neural networks with parallel distributed genetic programming. Technical Report CSRP-96-14, University of Birmingham, School of Computer Science, August 1996a. URL ftp://ftp.cs.bham.ac.uk/
pub/tech-reports/1996/CSRP-96-14.ps.gz. Presented at 3rd International Conference on Artificial Neural Networks and Genetic Algorithms, ICANNGA97.
GPBiB
R. Poli. Genetic programming for image analysis. In J. R. Koza, et al., editors, Genetic Programming 1996: Proceedings of the First Annual Conference, pages 363
368, Stanford University, CA, USA, 2831 July 1996b. MIT Press. URL http:
//cswww.essex.ac.uk/staff/rpoli/papers/Poli-GP1996.pdf.
GPBiB
R. Poli. Parallel distributed genetic programming. In D. Corne, et al., editors, New Ideas
in Optimization, Advanced Topics in Computer Science, chapter 27, pages 403431.
McGraw-Hill, Maidenhead, Berkshire, England, 1999a. ISBN 0-07-709506-5. URL
http://citeseer.ist.psu.edu/328504.html.
GPBiB
R. Poli. Sub-machine-code GP: New results and extensions. In R. Poli, et al., editors,
Genetic Programming, Proceedings of EuroGP99, volume 1598 of LNCS, pages 65
82, Goteborg, Sweden, 26-27 May 1999b. Springer-Verlag. ISBN 3-540-65899-8. URL
http://www.cs.essex.ac.uk/staff/poli/papers/Poli-EUROGP1999.pdf.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=27)

1.5

80
1
100
0.5

60
Fitness

Generation 27
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

206

BIBLIOGRAPHY

R. Poli. Hyperschema theory for GP with one-point crossover, building blocks, and some
new results in GA theory. In R. Poli, et al., editors, Genetic Programming, Proceedings
of EuroGP2000, volume 1802 of LNCS, pages 163180, Edinburgh, 15-16 April 2000a.
Springer-Verlag. ISBN 3-540-67339-3. URL http://www.springerlink.com/openurl.
asp?genre=article&issn=0302-9743&volume=1802&spage=163.
GPBiB
R. Poli. Exact schema theorem and effective fitness for GP with one-point crossover. In
D. Whitley, et al., editors, Proceedings of the Genetic and Evolutionary Computation
Conference, pages 469476, Las Vegas, July 2000b. Morgan Kaufmann.
R. Poli. Exact schema theory for genetic programming and variable-length genetic algorithms with one-point crossover. Genetic Programming and Evolvable Machines, 2(2):
123163, 2001a.
R. Poli. General schema theory for genetic programming with subtree-swapping crossover.
In Genetic Programming, Proceedings of EuroGP 2001, LNCS, Milan, 18-20 April
2001b. Springer-Verlag.
R. Poli. A simple but theoretically-motivated method to control bloat in genetic programming. In C. Ryan, et al., editors, Genetic Programming, Proceedings of the 6th
European Conference, EuroGP 2003, LNCS, pages 211223, Essex, UK, 14-16 April
2003. Springer-Verlag.
R. Poli. Tournament selection, iterated coupon-collection problem, and backwardchaining evolutionary algorithms. In A. H. Wright, et al., editors, Foundations of Genetic Algorithms 8, volume 3469 of Lecture Notes in Computer Science, pages 132155,
Aizu-Wakamatsu City, Japan, 5-9 January 2005. Springer-Verlag. ISBN 3-540-27237-2.
URL http://www.cs.essex.ac.uk/staff/rpoli/papers/foga2005_Poli.pdf. GPBiB
R. Poli, C. Di Chio, and W. B. Langdon. Exploring extended particle swarms: a genetic
programming approach. In H.-G. Beyer, et al., editors, GECCO 2005: Proceedings of
the 2005 conference on Genetic and evolutionary computation, volume 1, pages 169
176, Washington DC, USA, 25-29 June 2005. ACM Press. ISBN 1-59593-010-8. URL
http://www.cs.essex.ac.uk/staff/poli/papers/geccopso2005.pdf.
GPBiB
R. Poli and W. B. Langdon. A new schema theory for genetic programming with one-point
crossover and point mutation. In J. R. Koza, et al., editors, Genetic Programming 1997:
Proceedings of the Second Annual Conference, pages 278285, Stanford University,
CA, USA, 13-16 July 1997. Morgan Kaufmann. URL http://citeseer.ist.psu.edu/
327495.html.
GPBiB
R. Poli and W. B. Langdon. Schema theory for genetic programming with one-point
crossover and point mutation. Evolutionary Computation, 6(3):231252, 1998a. URL
http://cswww.essex.ac.uk/staff/poli/papers/Poli-ECJ1998.pdf.
GPBiB
R. Poli and W. B. Langdon. On the search properties of different crossover operators
in genetic programming. In J. R. Koza, et al., editors, Genetic Programming 1998:
Proceedings of the Third Annual Conference, pages 293301, University of Wisconsin,
Madison, Wisconsin, USA, 22-25 July 1998b. Morgan Kaufmann. ISBN 1-55860-548-7.
URL http://www.cs.essex.ac.uk/staff/poli/papers/Poli-GP1998.pdf.
GPBiB
R. Poli and W. B. Langdon. Sub-machine-code genetic programming. In L. Spector,
et al., editors, Advances in Genetic Programming 3, chapter 13, pages 301323. MIT
Press, Cambridge, MA, USA, June 1999. ISBN 0-262-19423-6. URL http://cswww.
essex.ac.uk/staff/rpoli/papers/Poli-AIGP3-1999.pdf.
GPBiB

BIBLIOGRAPHY

207

R. Poli and W. B. Langdon. Running genetic programming backward. In R. L. Riolo,

et al., editors, Genetic Programming Theory and Practice. Kluwer, 2005a.
R. Poli and W. B. Langdon. Running genetic programming backward. In T. Yu, et al.,
editors, Genetic Programming Theory and Practice III, volume 9 of Genetic Programming, chapter 9, pages 125140. Springer, Ann Arbor, 12-14 May 2005b. ISBN
0-387-28110-X. URL http://www.cs.essex.ac.uk/staff/poli/papers/GPTP2005.pdf.
GPBiB
R. Poli and W. B. Langdon. Backward-chaining evolutionary algorithms. Artificial Intelligence, 170(11):953982, August 2006a. URL http://www.cs.essex.ac.uk/staff/
poli/papers/aijournal2006.pdf.
GPBiB
R. Poli and W. B. Langdon. Efficient markov chain model of machine code program
execution and halting. In R. L. Riolo, et al., editors, Genetic Programming Theory
and Practice IV, volume 5 of Genetic and Evolutionary Computation, chapter 13.
Springer, Ann Arbor, 11-13 May 2006b. ISBN 0-387-33375-4. URL http://www.cs.
essex.ac.uk/staff/poli/papers/GPTP2006.pdf.
GPBiB
R. Poli, W. B. Langdon, and S. Dignum. On the limiting distribution of program sizes
in tree-based genetic programming. In M. Ebner, et al., editors, Proceedings of the
10th European Conference on Genetic Programming, volume 4445 of Lecture Notes in
Computer Science, pages 193204, Valencia, Spain, 11 - 13 April 2007. Springer. ISBN
3-540-71602-5.
GPBiB
R. Poli, W. B. Langdon, and O. Holland. Extending particle swarm optimisation via genetic programming. In M. Keijzer, et al., editors, Proceedings of the 8th European Conference on Genetic Programming, volume 3447 of Lecture Notes in Computer Science,
pages 291300, Lausanne, Switzerland, 30 March - 1 April 2005. Springer. ISBN 3-54025436-6. URL http://www.cs.essex.ac.uk/staff/poli/papers/eurogpPSO2005.pdf.
GPBiB
R.

Poli, W. B. Langdon, and N. F. McPhee.

A field guide to genetic programming.
Published via http://lulu.com and freely available at
http://www.gp-field-guide.org.uk, 2008. (With contributions by J. R. Koza). GPBiB

R. Poli and N. F. McPhee. A linear estimation-of-distribution GP system. In Proceedings

of EuroGP 2008, 2008a.
GPBiB
R. Poli and N. F. McPhee. Covariant parsimony pressure in genetic programming. Technical Report CES-480, Department of Computing and Electronic Systems, University
of Essex, January 2008b.
R. Poli and N. F. McPhee. Exact schema theorems for GP with one-point and standard crossover operating on linear structures and their application to the study of
the evolution of size. In J. F. Miller, et al., editors, Genetic Programming, Proceedings of EuroGP2001, volume 2038 of LNCS, pages 126142, Lake Como, Italy, 18-20
April 2001. Springer-Verlag. ISBN 3-540-41899-7. URL http://www.springerlink.
com/openurl.asp?genre=article&issn=0302-9743&volume=2038&spage=126. GPBiB
R. Poli and N. F. McPhee. General schema theory for genetic programming with subtreeswapping crossover: Part I. Evolutionary Computation, 11(1):5366, March 2003a.
URL http://cswww.essex.ac.uk/staff/rpoli/papers/ecj2003partI.pdf.
GPBiB
R. Poli and N. F. McPhee. General schema theory for genetic programming with subtreeswapping crossover: Part II. Evolutionary Computation, 11(2):169206, June 2003b.
URL http://cswww.essex.ac.uk/staff/rpoli/papers/ecj2003partII.pdf.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=29)

1.5

80
1
100
0.5

60
Fitness

Generation 29
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

208

BIBLIOGRAPHY

R. Poli, N. F. McPhee, and J. E. Rowe. Exact schema theory and markov chain models for genetic programming and variable-length genetic algorithms with homologous
crossover. Genetic Programming and Evolvable Machines, 5(1):3170, March 2004.
ISSN 1389-2576. URL http://cswww.essex.ac.uk/staff/rpoli/papers/GPEM2004.
pdf.
GPBiB
R. Poli and J. Page. Solving high-order boolean parity problems with smooth uniform
crossover, sub-machine code GP and demes. Genetic Programming and Evolvable
Machines, 1(1/2):3756, April 2000. ISSN 1389-2576. URL http://citeseer.ist.
psu.edu/335584.html.
GPBiB
R. Poli, J. Page, and W. B. Langdon. Smooth uniform crossover, sub-machine code GP
and demes: A recipe for solving high-order boolean parity problems. In W. Banzhaf,
et al., editors, Proceedings of the Genetic and Evolutionary Computation Conference, volume 2, pages 11621169, Orlando, Florida, USA, 13-17 July 1999. Morgan Kaufmann. ISBN 1-55860-611-4. URL http://www.cs.bham.ac.uk/~wbl/biblio/
gecco1999/GP-466.pdf.
GPBiB
R. Poli, J. E. Rowe, and N. F. McPhee. Markov chain models for GP and variable-length
GAs with homologous crossover. In L. Spector, et al., editors, Proceedings of the
Genetic and Evolutionary Computation Conference (GECCO-2001), pages 112119,
San Francisco, California, USA, 7-11 July 2001. Morgan Kaufmann. ISBN 1-55860774-9. URL http://www.cs.bham.ac.uk/~wbl/biblio/gecco2001/d01.pdf.
GPBiB
R. Poli, J. Woodward, and E. K. Burke. A histogram-matching approach to the evolution of bin-packing strategies. In Proceedings of the IEEE Congress on Evolutionary
Computation, Singapore, 2007. accepted.
R. Poli, W. B. Langdon, M. Schoenauer, T. Fogarty, and W. Banzhaf, editors. Late Breaking Papers at EuroGP98: the First European Workshop on Genetic Programming,
Paris, France, 14-15 April 1998. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/
ftp/papers/csrp-98-10.pdf.
GPBiB
R. Poli, P. Nordin, W. B. Langdon, and T. C. Fogarty, editors. Genetic Programming, Proceedings of EuroGP99, volume 1598 of LNCS, Goteborg, Sweden, 26-27
May 1999. Springer-Verlag. URL http://www.springerlink.com/openurl.asp?genre=
article&issn=0302-9743&volume=1598.
GPBiB
R. Poli, W. Banzhaf, W. B. Langdon, J. F. Miller, P. Nordin, and T. C. Fogarty, editors. Genetic Programming, Proceedings of EuroGP2000, volume 1802 of LNCS,
Edinburgh, 15-16 April 2000. Springer-Verlag. ISBN 3-540-67339-3.
GPBiB
E. Popovici and K. De Jong. The effects of interaction frequency on the optimization
performance of cooperative coevolution. In M. Keijzer, et al., editors, GECCO 2006:
Proceedings of the 8th annual conference on Genetic and evolutionary computation,
volume 1, pages 353360, Seattle, Washington, USA, 8-12 July 2006. ACM Press. ISBN
1-59593-186-4. URL http://www.cs.bham.ac.uk/~wbl/biblio/gecco2006/docs/p353.
pdf.
M. A. Potter. The Design and Analysis of a Computational Model of Cooperative Coevolution. PhD thesis, George Mason University, Washington, DC, spring 1997. URL
http://www.cs.gmu.edu/~mpotter/dissertation.html.
S. Priesterjahn, O. Kramer, A. Weimer, and A. Goebels. Evolution of human-competitive
agents in modern computer games. In G. G. Yen, et al., editors, Proceedings of the
2006 IEEE Congress on Evolutionary Computation, pages 777784, Vancouver, BC,
Canada, 16-21 July 2006. IEEE Press. ISBN 0-7803-9487-9. URL http://ieeexplore.
ieee.org/servlet/opac?punumber=11108.

BIBLIOGRAPHY

209

A. Pr
ugel-Bennett and J. L. Shapiro. An analysis of genetic algorithms using statistical
mechanics. Physical Review Letters, 72:13051309, 1994.
J. C. F. Pujol. Evolution of Artificial Neural Networks Using a Two-dimensional Representation. PhD thesis, School of Computer Science, University of Birmingham, UK,
April 1999.
GPBiB
J. C. F. Pujol and R. Poli. Evolution of the topology and the weights of neural networks
using genetic programming with a dual representation. Technical Report CSRP-97-7,
University of Birmingham, School of Computer Science, February 1997. URL ftp:
//ftp.cs.bham.ac.uk/pub/tech-reports/1997/CSRP-97-07.ps.gz.
GPBiB
B. Punch and D. Zongker. lil-gp Genetic Programming System, 1998. URL http://
garage.cse.msu.edu/software/lil-gp/index.html.
M. I. Quintana, R. Poli, and E. Claridge. On two approaches to image processing algorithm design for binary images using GP. In G. R. Raidl, et al., editors, Applications
of Evolutionary Computing, EvoWorkshops2003: EvoBIO, EvoCOP, EvoIASP, EvoMUSART, EvoROB, EvoSTIM, volume 2611 of LNCS, pages 422431, University of
Essex, England, UK, 14-16 April 2003. Springer-Verlag.
GPBiB
M. I. Quintana, R. Poli, and E. Claridge. Morphological algorithm design for binary
images using genetic programming. Genetic Programming and Evolvable Machines,
7(1):81102, March 2006. ISSN 1389-2576. URL http://cswww.essex.ac.uk/staff/
rpoli/papers/gpem2005.pdf.
GPBiB
A. Ratle and M. Sebag. Genetic programming and domain knowledge: Beyond the limitations of grammar-guided machine discovery. In M. Schoenauer, et al., editors, Parallel
Problem Solving from Nature - PPSN VI 6th International Conference, volume 1917
of LNCS, pages 211220, Paris, France, 16-20 September 2000. Springer Verlag. URL
http://www.lri.fr/~sebag/REF/PPSN00.ps.
GPBiB
A. Ratle and M. Sebag. Avoiding the bloat with probabilistic grammar-guided genetic
programming. In P. Collet, et al., editors, Artificial Evolution 5th International
Conference, Evolution Artificielle, EA 2001, volume 2310 of LNCS, pages 255266,
Creusot, France, October 29-31 2001. Springer Verlag. ISBN 3-540-43544-1. URL http:
//link.springer.de/link/service/series/0558/papers/2310/23100255.pdf. GPBiB
J. Reggia, M. Tagamets, J. Contreras-Vidal, D. Jacobs, S. Weems, W. Naqvi, R. Winder,
T. Chabuk, J. Jung, and C. Yang. Development of a large-scale integrated neurocognitive architecture - part 2: Design and architecture. Technical Report TRCS-4827, UMIACS-TR-2006-43, University of Maryland, USA, October 2006. URL
https://drum.umd.edu/dspace/bitstream/1903/3957/1/MarylandPart2.pdf. GPBiB
E. N. Regolin and A. T. R. Pozo. Bayesian automatic programming. In M. Keijzer, et al., editors, Proceedings of the 8th European Conference on Genetic Programming, volume 3447 of Lecture Notes in Computer Science, pages 3849,
Lausanne, Switzerland, 30 March - 1 April 2005. Springer. ISBN 3-540-254366.
URL http://springerlink.metapress.com/openurl.asp?genre=article&issn=
0302-9743&volume=3447&spage=38.
GPBiB
D. M. Reif, B. C. White, and J. H. Moore. Integrated analysis of genetic, genomic, and
proteomic data. Expert Review of Proteomics, 1(1):6775, 2004. ISSN 1473-7159. URL
http://www.future-drugs.com/doi/abs/10.1586/14789450.1.1.67.
GPBiB
C. W. Reynolds. Flocks, herds, and schools: A distributed behavioral model. SIGGRAPH
Computer Graphics, 21(4):2534, July 1987. ISSN 0097-8930. URL http://www.red3d.
com/cwr/papers/1987/boids.html.
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=36)

1.5

80
1
100
0.5

60
Fitness

Generation 36
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

210

BIBLIOGRAPHY

R. L. Riolo and B. Worzel. Genetic Programming Theory and Practice, volume 6 of

Genetic Programming. Kluwer, Boston, MA, USA, 2003. ISBN 1-4020-7581-2. URL
http://www.wkap.nl/prod/b/1-4020-7581-2. Series Editor - John Koza.
GPBiB
R. L. Riolo, T. Soule, and B. Worzel, editors. Genetic Programming Theory and Practice
IV, volume 5 of Genetic and Evolutionary Computation, Ann Arbor, 11-13 May 2007a.
Springer. ISBN 0-387-33375-4. URL http://www.springer.com/west/home/computer/
foundations?SGWID=4-156-22-173660377-0.
GPBiB
R. L. Riolo, T. Soule, and B. Worzel, editors. Genetic Programming Theory and Practice
V, Genetic and Evolutionary Computation, Ann Arbor, 17-19 May 2007b. Springer.
GPBiB
M. D. Ritchie, A. A. Motsinger, W. S. Bush, C. S. Coffey, and J. H. Moore. Genetic
programming neural networks: A powerful bioinformatics tool for human genetics.
Applied Soft Computing, 7(1):471479, January 2007.
GPBiB
M. D. Ritchie, B. C. White, J. S. Parker, L. W. Hahn, and J. H. Moore. Optimization of
neural network architecture using genetic programming improves detection and modeling of gene-gene interactions in studies of human diseases. BMC Bioinformatics, 4
(28), 7 July 2003. URL http://www.biomedcentral.com/1471-2105/4/28.
GPBiB
D. Rivero, J. R. R. nal, J. Dorado, and A. Pazos. Using genetic programming for character
discrimination in damaged documents. In G. R. Raidl, et al., editors, Applications
of Evolutionary Computing, EvoWorkshops2004: EvoBIO, EvoCOMNET, EvoHOT,
EvoIASP, EvoMUSART, EvoSTOC, volume 3005 of LNCS, pages 349358, Coimbra,
Portugal, 5-7 April 2004. Springer Verlag. ISBN 3-540-21378-3.
GPBiB
RML Technologies. Discipulus Genetic-Programming Software, 1998-2007. URL http:
//www.rmltech.com/.
A. Robinson and L. Spector. Using genetic programming with multiple data types and
automatic modularization to evolve decentralized and coordinated navigation in multiagent systems. In E. Cant
u-Paz, editor, Late Breaking Papers at the Genetic and Evolutionary Computation Conference (GECCO-2002), pages 391396, New York, NY,
July 2002. AAAI.
GPBiB
K. Rodriguez-Vazquez, C. M. Fonseca, and P. J. Fleming. Multiobjective genetic programming: A nonlinear system identification application. In J. R. Koza, editor, Late
Breaking Papers at the 1997 Genetic Programming Conference, pages 207212, Stanford University, CA, USA, 1316 July 1997. Stanford Bookstore. ISBN 0-18-206995-8.
URL http://www.lania.mx/~ccoello/EMOO/katya97.ps.gz.
GPBiB
K. Rodriguez-Vazquez, C. M. Fonseca, and P. J. Fleming. Identifying the structure of
nonlinear dynamic systems using multiobjective genetic programming. IEEE Transactions on Systems, Man and Cybernetics, Part A, 34(4):531545, July 2004. ISSN
1083-4427.
GPBiB
J. P. Rosca. Generality versus size in genetic programming. In J. R. Koza, et al., editors,
Genetic Programming 1996: Proceedings of the First Annual Conference, pages 381
387, Stanford University, CA, USA, 2831 July 1996. MIT Press. URL ftp://ftp.cs.
rochester.edu/pub/u/rosca/gp/96.gp.ps.gz.
GPBiB
J. P. Rosca. Analysis of complexity drift in genetic programming. In J. R. Koza, et al.,
editors, Genetic Programming 1997: Proceedings of the Second Annual Conference,
pages 286294, Stanford University, CA, USA, 13-16 July 1997. Morgan Kaufmann.
URL ftp://ftp.cs.rochester.edu/pub/u/rosca/gp/97.gp.ps.gz.
GPBiB

BIBLIOGRAPHY

211

J. P. Rosca. A probabilistic model of size drift. In R. L. Riolo and B. Worzel, editors,

Genetic Programming Theory and Practice, chapter 8, pages 119136. Kluwer, 2003.
GPBiB
J. P. Rosca and D. H. Ballard. Discovery of subroutines in genetic programming. In
P. J. Angeline and K. E. Kinnear, Jr., editors, Advances in Genetic Programming 2,
chapter 9, pages 177202. MIT Press, Cambridge, MA, USA, 1996a. ISBN 0-26201158-1. URL ftp://ftp.cs.rochester.edu/pub/u/rosca/gp/96.aigp2.dsgp.ps.gz.
GPBiB
J. P. Rosca and D. H. Ballard. Complexity drift in evolutionary computation with tree
representations. Technical Report NRL5, University of Rochester, Computer Science
Department, Rochester, NY, USA, December 1996b. URL ftp://ftp.cs.rochester.
edu/pub/u/rosca/gp/96.drift.ps.gz.
GPBiB
J. P. Rosca and D. H. Ballard. Rooted-tree schemata in genetic programming. In
L. Spector, et al., editors, Advances in Genetic Programming 3, chapter 11, pages
243271. MIT Press, Cambridge, MA, USA, June 1999. ISBN 0-262-19423-6. URL
http://www.cs.bham.ac.uk/~wbl/aigp3/ch11.pdf.
GPBiB
B. J. Ross, A. G. Gualtieri, F. Fueten, and P. Budkewitsch. Hyperspectral image analysis
using genetic programming. Applied Soft Computing, 5(2):147156, January 2005.
URL http://www.cosc.brocku.ca/~bross/research/gp_hyper.pdf.
GPBiB
B. J. Ross and H. Zhu. Procedural texture evolution using multiobjective optimization.
New Generation Computing, 22(3):271293, 2004. URL http://www.cosc.brocku.ca/
files/downloads/research/cs0218.pdf.
GPBiB
R. Rossi, V. Liberali, and A. G. B. Tettamanzi. An application of genetic programming
to electronic design automation: from frequency specifications to VHDL code. In
R. Roy, et al., editors, Soft Computing and Industry Recent Applications, pages 809
820. Springer-Verlag, 1024 September 2001. ISBN 1-85233-539-4. URL http://mago.
crema.unimi.it/pub/RossiLiberaliTettamanzi2001.pdf. Published 2002.
GPBiB
F. Rothlauf. Representations for genetic and evolutionary algorithms. Springer-Verlag,
pub-SV:adr, second edition, 2006. ISBN 3-540-25059-X. URL http://download-ebook.
org/index.php?target=desc&ebookid=5771. First published 2002, 2nd edition available electronically.
C. Ryan. Pygmies and civil servants. In K. E. Kinnear, Jr., editor, Advances in Genetic
Programming, chapter 11, pages 243263. MIT Press, 1994. URL http://cognet.mit.
edu/library/books/view?isbn=0262111888.
GPBiB
C. Ryan. Reducing Premature Convergence in Evolutionary Algorithms. PhD thesis,
University College, Cork, Ireland, 2 July 1996. URL http://citeseer.ist.psu.edu/
cache/papers/cs/6401/ftp:zSzzSzodyssey.ucc.iezSzpubzSzgeneticzSzthesis.pdf/
ryan96reducing.pdf.
GPBiB
C. Ryan. Automatic Re-engineering of Software Using Genetic Programming, volume 2
of Genetic Programming. Kluwer Academic Publishers, 1 November 1999. ISBN 07923-8653-1. URL http://www.wkap.nl/book.htm/0-7923-8653-1.
GPBiB
C. Ryan, J. J. Collins, and M. ONeill. Grammatical evolution: Evolving programs for an
arbitrary language. In W. Banzhaf, et al., editors, Proceedings of the First European
Workshop on Genetic Programming, volume 1391 of LNCS, pages 8395, Paris, 1415 April 1998. Springer-Verlag. ISBN 3-540-64360-5. URL http://www.lania.mx/
~ccoello/eurogp98.ps.gz.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=43)

1.5

80
1
100
0.5

60
Fitness

Generation 43
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

212

BIBLIOGRAPHY

C. Ryan, T. Soule, M. Keijzer, E. P. K. Tsang, R. Poli, and E. Costa, editors. Genetic Programming, Proceedings of the 6th European Conference, EuroGP 2003, volume 2610 of LNCS, Essex, UK, 14-16 April 2003. Springer-Verlag. ISBN 3-54000971-X. URL http://www.springerlink.com/openurl.asp?genre=article&issn=
0302-9743&volume=2610.
GPBiB
R. P. Salustowicz and J. Schmidhuber. Probabilistic incremental program evolution. Evolutionary Computation, 5(2):123141, 1997. URL ftp://ftp.idsia.ch/pub/rafal/
PIPE.ps.gz.
GPBiB
R. P. Salustowicz, M. A. Wiering, and J. Schmidhuber. Learning team strategies: Soccer
case studies. Machine Learning, 33(2-3):263282, 12 November 1998. ISSN 0885-6125.
URL ftp://ftp.idsia.ch/pub/rafal/soccer.ps.gz.
R. P. Salustowicz and J. Schmidhuber. From probabilities to programs with probabilistic
incremental program evolution. In D. Corne, et al., editors, New Ideas in Optimization, Advanced Topics in Computer Science, chapter 28, pages 433450. McGraw-Hill,
Maidenhead, Berkshire, England, 1999. ISBN 0-07-709506-5.
GPBiB
A. L. Samuel. AI, where it has been and where it is going. In IJCAI, pages 11521157,
1983.
A. Sarafopoulos. Automatic generation of affine IFS and strongly typed genetic programming. In R. Poli, et al., editors, Genetic Programming, Proceedings of EuroGP99, volume 1598 of LNCS, pages 149160, Goteborg, Sweden, 26-27 May 1999.
Springer-Verlag. ISBN 3-540-65899-8. URL http://www.springerlink.com/openurl.
asp?genre=article&issn=0302-9743&volume=1598&spage=149.
GPBiB
K. Sastry and D. E. Goldberg. Probabilistic model building and competent genetic programming. In R. L. Riolo and B. Worzel, editors, Genetic Programming Theory and
Practise, chapter 13, pages 205220. Kluwer, 2003. ISBN 1-4020-7581-2.
GPBiB
M. D. Schmidt and H. Lipson. Co-evolving fitness predictors for accelerating and reducing
evaluations. In R. L. Riolo, et al., editors, Genetic Programming Theory and Practice
IV, volume 5 of Genetic and Evolutionary Computation, chapter 17, pages . Springer,
Ann Arbor, 11-13 May 2006. ISBN 0-387-33375-4.
GPBiB
F. Schmiedle, N. Drechsler, D. Grosse, and R. Drechsler. Priorities in multi-objective
optimization for genetic programming. In L. Spector, et al., editors, Proceedings of the
Genetic and Evolutionary Computation Conference (GECCO-2001), pages 129136,
San Francisco, California, USA, 7-11 July 2001. Morgan Kaufmann. ISBN 1-55860774-9. URL http://www.cs.bham.ac.uk/~wbl/biblio/gecco2001/d01.pdf.
GPBiB
M. Schoenauer, B. Lamy, and F. Jouve. Identification of mechanical behaviour by genetic programming part II: Energy formulation. Technical report, Ecole Polytechnique,
91128 Palaiseau, France, 1995.
GPBiB
M. Schoenauer and M. Sebag. Using domain knowledge in evolutionary system identification. In K. C. Giannakoglou, et al., editors, Evolutionary Methods for Design,
Optimization and Control with Applications to Industrial Problems, Athens, 19-21
September 2001. URL http://arxiv.org/abs/cs/0602021.
GPBiB
M. Schoenauer, M. Sebag, F. Jouve, B. Lamy, and H. Maitournam. Evolutionary identification of macro-mechanical models.
In P. J. Angeline and K. E.
Kinnear, Jr., editors, Advances in Genetic Programming 2, chapter 23, pages
467488. MIT Press, Cambridge, MA, USA, 1996. ISBN 0-262-01158-1. URL
http://citeseer.ist.psu.edu/cache/papers/cs/902/http:zSzzSzwww.eeaax.

BIBLIOGRAPHY

213

polytechnique.frzSzpaperszSzmarczSzAGP2.pdf/schoenauer96evolutionary.pdf.
GPBiB
M. Schoenauer, K. Deb, G. Rudolph, X. Yao, E. Lutton, J. J. Merelo, and H.-P. Schwefel,
editors. Parallel Problem Solving from Nature - PPSN VI 6th International Conference, volume 1917 of LNCS, Paris, France, 16-20 September 2000. Springer Verlag.
ISBN 3-540-41056-2. URL http://www.springer.de/cgi-bin/search_book.pl?isbn=
3-540-41056-2.
GPBiB
D. P. Searson, G. A. Montague, and M. J. Willis. Evolutionary design of process controllers. In In Proceedings of the 1998 United Kingdom Automatic Control Council International Conference on Control (UKACC International Conference on Control 98), volume 455 of IEE Conference Publications, University of Wales, Swansea,
UK, 1-4 September 1998. Institution of Electrical Engineers (IEE). URL http:
//www.staff.ncl.ac.uk/d.p.searson/docs/Searsoncontrol98.pdf.
GPBiB
L. Sekanina. Evolvable Components: From Theory to Hardware Implementations. Natural
Computing. Springer-Verlag, 2003. ISBN 3-540-40377-9. URL http://www.fit.vutbr.
cz/~sekanina/ehw/books.html.en.
H.-S. Seok, K.-J. Lee, and B.-T. Zhang. An on-line learning method for objectlocating robots using genetic programming on evolvable hardware. In M. Sugisaka
and H. Tanaka, editors, Proceedings of the Fifth International Symposium on Artificial Life and Robotics, volume 1, pages 321324, Oita, Japan, 26-28 January 2000. URL
http://bi.snu.ac.kr/Publications/Conferences/International/AROB00.ps. GPBiB
C. Setzkorn. On The Use Of Multi-Objective Evolutionary Algorithms For Classification
Rule Induction. PhD thesis, University of Liverpool, UK, March 2005.
GPBiB
S. C. Shah and A. Kusiak. Data mining and genetic algorithm based gene/SNP selection.
Artificial Intelligence in Medicine, 31(3):183196, July 2004. URL http://www.icaen.
uiowa.edu/~ankusiak/Journal-papers/Gen_Shital.pdf.
GPBiB
Y. Shan, H. Abbass, R. I. McKay, and D. Essam. AntTAG: a further study. In R. Sarker
and B. McKay, editors, Proceedings of the Sixth Australia-Japan Joint Workshop on
Intelligent and Evolutionary Systems, Australian National University, Canberra, Australia, 30 November 2002.
GPBiB
Y. Shan, R. I. McKay, H. A. Abbass, and D. Essam. Program evolution with explicit
learning: a new framework for program automatic synthesis. In R. Sarker, et al.,
editors, Proceedings of the 2003 Congress on Evolutionary Computation CEC2003,
pages 16391646, Canberra, 8-12 December 2003. IEEE Press. ISBN 0-7803-7804-0.
URL http://citeseer.ist.psu.edu/560804.html.
GPBiB
Y. Shan, R. I. McKay, R. Baxter, H. Abbass, D. Essam, and N. X. Hoai. Grammar
model-based program evolution. In Proceedings of the 2004 IEEE Congress on Evolutionary Computation, pages 478485, Portland, Oregon, 20-23 June 2004. IEEE
Press. ISBN 0-7803-8515-2. URL http://sc.snu.ac.kr/courses/2006/fall/pg/aai/
GP/shan/scfgcec04.pdf.
GPBiB
Y. Shan, R. I. McKay, D. Essam, and H. A. Abbass. A survey of probabilistic model
building genetic programming. In M. Pelikan, et al., editors, Scalable Optimization
via Probabilistic Modeling: From Algorithms to Applications. Springer, 2006. ISBN
3-540-34953-7.
GPBiB
S. Sharabi and M. Sipper. GP-sumo: Using genetic programming to evolve sumobots.
Genetic Programming and Evolvable Machines, 7(3):211230, October 2006. ISSN
1389-2576.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=50)

1.5

80
1
100
0.5

60
Fitness

Generation 50
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

214

BIBLIOGRAPHY

K. C. Sharman and A. I. Esparcia-Alcazar. Genetic evolution of symbolic signal models.

In Proceedings of the Second International Conference on Natural Algorithms in Signal
Processing, NASP93, Essex University, UK, 15-16 November 1993. URL http://www.
iti.upv.es/~anna/papers/natalg93.ps.
GPBiB
K. C. Sharman, A. I. Esparcia Alcazar, and Y. Li. Evolving signal processing algorithms
by genetic programming. In A. M. S. Zalzala, editor, First International Conference on
Genetic Algorithms in Engineering Systems: Innovations and Applications, GALESIA, volume 414, pages 473480, Sheffield, UK, 12-14 September 1995. IEE. ISBN
0-85296-650-4. URL http://www.iti.upv.es/~anna/papers/galesi95.ps.
GPBiB
A. D. Shaw, M. K. Winson, A. M. Woodward, A. C. McGovern, H. M. Davey, N. Kaderbhai, D. Broadhurst, R. J. Gilbert, J. Taylor, E. M. Timmins, R. Goodacre, D. B. Kell,
B. K. Alsberg, and J. J. Rowland. Bioanalysis and biosensors for bioprocess monitoring rapid analysis of high-dimensional bioprocesses using multivariate spectroscopies
and advanced chemometrics. Advances in Biochemical Engineering/Biotechnology, 66:
83113, January 2000. ISSN 0724-6145. URL http://www.springerlink.com/link.
asp?id=t8b4ya0bl42jnjj3.
GPBiB
Y. Shichel, E. Ziserman, and M. Sipper. GP-robocode: Using genetic programming to
evolve robocode players. In M. Keijzer, et al., editors, Proceedings of the 8th European
Conference on Genetic Programming, volume 3447 of Lecture Notes in Computer Science, pages 143154, Lausanne, Switzerland, 30 March - 1 April 2005. Springer. ISBN
3-540-25436-6. URL http://www.cs.bgu.ac.il/~sipper/papabs/eurogprobo-final.
pdf.
GPBiB
H. Z. Si, T. Wang, K. J. Zhang, Z. D. Hu, and B. T. Fan. QSAR study of 1,4dihydropyridine calcium channel antagonists based on gene expression programming.
Bioorganic & Medicinal Chemistry, 14(14):48344841, 15 July 2006.
GPBiB
R. P. S. Sidhu, A. Mei, and V. K. Prasanna. Genetic programming using selfreconfigurable FPGAs. In 9th International Workshop on Field Programmable Logic
and Applications, volume 1673 of LNCS, pages 301312, Glasgow, UK, 30 August - 1
September 1998. Springer-Verlag. ISBN 3-540-66457-2. URL http://ceng.usc.edu/
~prasanna/papers/sidhuFPL99.ps.
GPBiB
E. V. Siegel. Competitively evolving decision trees against fixed training cases for natural
language processing. In K. E. Kinnear, Jr., editor, Advances in Genetic Programming,
chapter 19, pages 409423. MIT Press, 1994. URL http://www1.cs.columbia.edu/
nlp/papers/1994/siegel_94.pdf.
GPBiB
S. Silva and J. Almeida. Dynamic maximum tree depth. In E. Cant
u-Paz, et al., editors,
Genetic and Evolutionary Computation GECCO-2003, volume 2724 of LNCS, pages
17761787, Chicago, 12-16 July 2003. Springer-Verlag. ISBN 3-540-40603-4. URL
http://cisuc.dei.uc.pt/ecos/dlfile.php?fn=109_pub_27241776.pdf.
GPBiB
S. Silva and E. Costa. Dynamic limits for bloat control: Variations on size and depth. In
K. Deb, et al., editors, Genetic and Evolutionary Computation GECCO-2004, Part
II, volume 3103 of Lecture Notes in Computer Science, pages 666677, Seattle, WA,
USA, 26-30 June 2004. Springer-Verlag. ISBN 3-540-22343-6. URL http://cisuc.
dei.uc.pt/ecos/dlfile.php?fn=714_pub_31030666.pdf&idp=714.
GPBiB
S. Silva and E. Costa. Comparing tree depth limits and resource-limited GP. In D. Corne,
et al., editors, Proceedings of the 2005 IEEE Congress on Evolutionary Computation,
volume 1, pages 920927, Edinburgh, UK, 2-5 September 2005a. IEEE Press. ISBN
0-7803-9363-5.
GPBiB

BIBLIOGRAPHY

215

S. Silva and E. Costa. Resource-limited genetic programming: the dynamic approach.

In H.-G. Beyer, et al., editors, GECCO 2005: Proceedings of the 2005 conference on
Genetic and evolutionary computation, volume 2, pages 16731680, Washington DC,
USA, 25-29 June 2005b. ACM Press. ISBN 1-59593-010-8. URL http://www.cs.bham.
ac.uk/~wbl/biblio/gecco2005/docs/p1673.pdf.
GPBiB
S. Silva, P. J. N. Silva, and E. Costa. Resource-limited genetic programming: Replacing
tree depth limits. In B. Ribeiro, et al., editors, Adaptive and Natural Computing
Algorithms, Springer Computer Series, pages 243246, Coimbra, Portugal, 21-23 March
2005. Springer. ISBN 3-211-24934-6. URL http://www.lri.fr/~sebag/Examens/sara.
icannga05.pdf.
GPBiB
K. Sims. Artificial evolution for computer graphics. ACM Computer Graphics, 25(4):319
328, July 1991. URL http://delivery.acm.org/10.1145/130000/122752/p319-sims.
pdf. SIGGRAPH 91 Proceedings.
GPBiB
W. Smart and M. Zhang. Applying online gradient descent search to genetic programming for object recognition. In J. Hogan, et al., editors, CRPIT 04: Proceedings of the
second workshop on Australasian information security, Data Mining and Web Intelligence, and Software Internationalisation, volume 32 no. 7, pages 133138, Dunedin,
New Zealand, January 2004. Australian Computer Society, Inc. ISBN 1-920682-14-7.
URL http://crpit.com/confpapers/CRPITV32Smart.pdf.
GPBiB
T. Smith, S. Bullock, and J. Bird. Beyond fitness: Visualising evolution workshop
overview. In E. Bilotta, et al., editors, ALife VIII: Workshop proceedings, pages 100
102, 2002. URL http://www.alife.org/alife8/workshops/14.pdf.
T. Soule. Code Growth in Genetic Programming. PhD thesis, University of Idaho,
Moscow, Idaho, USA, 15 May 1998. URL http://www.cs.uidaho.edu/~tsoule/
research/the3.ps.
GPBiB
T. Soule. Cooperative evolution on the intertwined spirals problem. In C. Ryan,
et al., editors, Genetic Programming, Proceedings of EuroGP2003, volume 2610
of LNCS, pages 434442, Essex, 14-16 April 2003. Springer-Verlag. ISBN 3-54000971-X. URL http://www.springerlink.com/openurl.asp?genre=article&issn=
0302-9743&volume=2610&spage=434.
GPBiB
T. Soule and J. A. Foster. Removal bias: a new cause of code growth in tree based
evolutionary programming. In 1998 IEEE International Conference on Evolutionary
Computation, pages 781186, Anchorage, Alaska, USA, 5-9 May 1998a. IEEE Press.
URL http://citeseer.ist.psu.edu/313655.html.
GPBiB
T. Soule and J. A. Foster. Effects of code growth and parsimony pressure on populations
in genetic programming. Evolutionary Computation, 6(4):293309, Winter 1998b. URL
http://mitpress.mit.edu/journals/EVCO/Soule.pdf.
GPBiB
T. Soule and P. Komireddy. Orthogonal evolution of teams: A class of algorithms for
evolving teams with inversely correlated errors. In R. L. Riolo, et al., editors, Genetic
Programming Theory and Practice IV, volume 5 of Genetic and Evolutionary Computation, chapter 8, pages . Springer, Ann Arbor, 11-13 May 2006. ISBN 0-387-33375-4.
GPBiB
L. Spector. Autoconstructive evolution: Push, pushGP, and pushpop. In L. Spector,
et al., editors, Proceedings of the Genetic and Evolutionary Computation Conference
(GECCO-2001), pages 137146, San Francisco, California, USA, 7-11 July 2001. Morgan Kaufmann. ISBN 1-55860-774-9. URL http://hampshire.edu/lspector/pubs/
ace.pdf.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=57)

1.5

80
1
100
0.5

60
Fitness

Generation 57
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

216

BIBLIOGRAPHY

L. Spector. Adaptive populations of endogenously diversifying pushpop organisms are

reliably diverse. In R. K. Standish, et al., editors, Proceedings of Artificial Life VIII, the
8th International Conference on the Simulation and Synthesis of Living Systems, pages
142145, University of New South Wales, Sydney, NSW, Australia, 9th-13th December
2002. The MIT Press. URL http://hampshire.edu/lspector/pubs/spector-alife8.
pdf.
GPBiB
L. Spector. Automatic Quantum Computer Programming: A Genetic Programming Approach, volume 7 of Genetic Programming. Kluwer Academic Publishers, Boston/Dordrecht/New York/London, June 2004. ISBN 1-4020-7894-3. URL http://www.wkap.
nl/prod/b/1-4020-7894-3.
GPBiB
L. Spector and A. Alpern. Criticism, culture, and the automatic generation of artworks.
In Proceedings of Twelfth National Conference on Artificial Intelligence, pages 38,
Seattle, Washington, USA, 1994. AAAI Press/MIT Press.
GPBiB
L. Spector and A. Alpern. Induction and recapitulation of deep musical structure. In
Proceedings of International Joint Conference on Artificial Intelligence, IJCAI95
Workshop on Music and AI, Montreal, Quebec, Canada, 20-25 August 1995. URL
http://hampshire.edu/lspector/pubs/IJCAI95mus-toappear.ps.
GPBiB
L. Spector, H. Barnum, and H. J. Bernstein. Genetic programming for quantum computers. In J. R. Koza, et al., editors, Genetic Programming 1998: Proceedings of the
Third Annual Conference, pages 365373, University of Wisconsin, Madison, Wisconsin, USA, 22-25 July 1998. Morgan Kaufmann. ISBN 1-55860-548-7.
GPBiB
L. Spector, H. Barnum, H. J. Bernstein, and N. Swamy. Finding a better-than-classical
quantum AND/OR algorithm using genetic programming. In P. J. Angeline, et al., editors, Proceedings of the Congress on Evolutionary Computation, volume 3, pages 2239
2246, Mayflower Hotel, Washington D.C., USA, 6-9 July 1999. IEEE Press. ISBN 07803-5536-9 (softbound). URL http://hampshire.edu/~lasCCS/pubs/spector-cec99.
ps.
GPBiB
L. Spector and J. Klein. Multidimensional tags, cooperative populations, and genetic
programming. In R. L. Riolo, et al., editors, Genetic Programming Theory and Practice
IV, volume 5 of Genetic and Evolutionary Computation, chapter 15, pages . Springer,
Ann Arbor, 11-13 May 2006. ISBN 0-387-33375-4.
GPBiB
L. Spector, J. Klein, and M. Keijzer. The push3 execution stack and the evolution
of control. In H.-G. Beyer, et al., editors, GECCO 2005: Proceedings of the 2005
conference on Genetic and evolutionary computation, volume 2, pages 16891696,
Washington DC, USA, 25-29 June 2005a. ACM Press. ISBN 1-59593-010-8. URL
http://www.cs.bham.ac.uk/~wbl/biblio/gecco2005/docs/p1689.pdf.
GPBiB
L. Spector, J. Klein, C. Perry, and M. Feinstein. Emergence of collective behavior in
evolving populations of flying agents. Genetic Programming and Evolvable Machines,
6(1):111125, March 2005b. ISSN 1389-2576. URL http://hampshire.edu/lspector/
pubs/emergence-collective-GPEM.pdf.
GPBiB
L. Spector, W. B. Langdon, U.-M. OReilly, and P. J. Angeline, editors. Advances in
Genetic Programming 3. MIT Press, Cambridge, MA, USA, June 1999. ISBN 0-26219423-6. URL http://www.cs.bham.ac.uk/~wbl/aigp3.
GPBiB
L. Spector, E. D. Goodman, A. Wu, W. B. Langdon, H.-M. Voigt, M. Gen, S. Sen,
M. Dorigo, S. Pezeshk, M. H. Garzon, and E. Burke, editors. Proceedings of the
Genetic and Evolutionary Computation Conference, GECCO-2001, San Francisco,
California, USA, 7-11 July 2001. Morgan Kaufmann. ISBN 1-55860-774-9. URL http:
//www.cs.bham.ac.uk/~wbl/biblio/gecco2001/.
GPBiB

BIBLIOGRAPHY

217

J. Stender, editor. Parallel Genetic Algorithms: Theory and Applications. IOS press,
1993.
C. R. Stephens and H. Waelbroeck. Effective degrees of freedom in genetic algorithms
and the block hypothesis. In T. B
ack, editor, Proceedings of the Seventh International Conference on Genetic Algorithms (ICGA97), pages 3440, East Lansing, 1997.
Morgan Kaufmann.
C. R. Stephens and H. Waelbroeck. Schemata evolution and building blocks. Evolutionary
Computation, 7(2):109124, 1999.
T. Sterling. Beowulf-class clustered computing: Harnessing the power of parallelism in a
pile of PCs. In J. R. Koza, et al., editors, Genetic Programming 1998: Proceedings of
the Third Annual Conference, page 883, University of Wisconsin, Madison, Wisconsin,
USA, 22-25 July 1998. Morgan Kaufmann. ISBN 1-55860-548-7. Invited talk.
A. Stoica, J. Lohn, and D. Keymeulen, editors. The First NASA/DoD Workshop on
Evolvable Hardware, Pasadena, California, 19-21 July 1999. IEEE Computer Society.
URL http://cism.jpl.nasa.gov/ehw/events/nasa_eh/.
W. A. Tackett.
Genetic generation of dendritic trees for image classification.
In Proceedings of WCNN93, pages IV 646649. IEEE Press, July
1993. URL http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/ftp.io.com/papers/GP.
feature.discovery.ps.Z.
GPBiB
H. Takagi. Interactive evolutionary computation: Fusion of the capabilities of EC optimization and human evaluation. Proceedings of the IEEE, 89(9):12751296, September
2001. ISSN 0018-9219. Invited Paper.
GPBiB
I. Tanev, T. Uozumi, and D. Akhmetov. Component object based single system image for
dependable implementation of genetic programming on clusters. Cluster Computing
Journal, 7(4):347356, October 2004. ISSN 1386-7857 (Paper) 1573-7543 (Online).
URL http://www.kluweronline.com/issn/1386-7857.
GPBiB
J. Taylor, R. Goodacre, W. G. Wade, J. J. Rowland, and D. B. Kell. The deconvolution
of pyrolysis mass spectra using genetic programming: application to the identification
of some eubacterium species. FEMS Microbiology Letters, 160:237246, 1998. GPBiB
A. Teller. Genetic programming, indexed memory, the halting problem, and other curiosities. In Proceedings of the 7th annual Florida Artificial Intelligence Research
Symposium, pages 270274, Pensacola, Florida, USA, May 1994. IEEE Press. URL
http://www.cs.cmu.edu/afs/cs/usr/astro/public/papers/Curiosities.ps. GPBiB
A. Teller. Evolving programmers: The co-evolution of intelligent recombination operators.
In P. J. Angeline and K. E. Kinnear, Jr., editors, Advances in Genetic Programming 2,
chapter 3, pages 4568. MIT Press, Cambridge, MA, USA, 1996. ISBN 0-262-01158-1.
URL http://www.cs.cmu.edu/afs/cs/usr/astro/public/papers/AiGPII.ps. GPBiB
A. Teller and D. Andre. Automatically choosing the number of fitness cases: The rational
allocation of trials. In J. R. Koza, et al., editors, Genetic Programming 1997: Proceedings of the Second Annual Conference, pages 321328, Stanford University, CA,
USA, 13-16 July 1997. Morgan Kaufmann. URL http://www.cs.cmu.edu/afs/cs/usr/
astro/public/papers/GR.ps.
GPBiB
A. Teredesai and V. Govindaraju. GP-based secondary classifiers. Pattern Recognition,
38(4):505512, April 2005.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=64)

1.5

80
1
100
0.5

60
Fitness

Generation 64
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

218

BIBLIOGRAPHY

J. P. Theiler, N. R. Harvey, S. P. Brumby, J. J. Szymanski, S. Alferink, S. J. Perkins,

R. B. Porter, and J. J. Bloch. Evolving retrieval algorithms with a genetic programming scheme. In M. R. Descour and S. S. Shen, editors, Proceedings of SPIE 3753
Imaging Spectrometry V, pages 416425, 1999. URL http://public.lanl.gov/jt/
Papers/ga-spie.ps.
GPBiB
D. Thierens, H.-G. Beyer, J. Bongard, J. Branke, J. A. Clark, D. Cliff, C. B. Congdon,
K. Deb, B. Doerr, T. Kovacs, S. Kumar, J. F. Miller, J. Moore, F. Neumann, M. Pelikan, R. Poli, K. Sastry, K. O. Stanley, T. Stutzle, R. A. Watson, and I. Wegener,
editors. GECCO 2007: Proceedings of the 9th annual conference on Genetic and
evolutionary computation, London, UK, 7-11 July 2007. ACM Press.
GPBiB
S. Thompson. Type theory and functional programming. Addison Wesley Longman Publishing Co., Inc., Redwood City, CA, USA, 1991. ISBN 0-201-41667-0.
P. M. Todd and G. M. Werner. Frankensteinian approaches to evolutionary music composition. In N. Griffith and P. M. Todd, editors, Musical Networks: Parallel Distributed
Perception and Performance, pages 313340. MIT Press, 1999. ISBN 0-262-071819. URL http://www-abc.mpib-berlin.mpg.de/users/ptodd/publications/99evmus/
99evmus.pdf.
GPBiB
M. Tomassini, L. Luthi, M. Giacobini, and W. B. Langdon. The structure of the genetic
programming collaboration network. Genetic Programming and Evolvable Machines,
8(1):97103, March 2007.
GPBiB
L. Trujillo and G. Olague. Using evolution to learn how to perform interest point detection. In X. Y. T. et al., editor, ICPR 2006 18th International Conference on
Pattern Recognition, volume 1, pages 211214. IEEE, 20-24 August 2006a. URL http:
//www.genetic-programming.org/hc2006/Olague-Paper-2-ICPR-2006.pdf.
GPBiB
L. Trujillo and G. Olague. Synthesis of interest point detectors through genetic programming. In M. Keijzer, et al., editors, GECCO 2006: Proceedings of the 8th annual conference on Genetic and evolutionary computation, volume 1, pages 887894,
Seattle, Washington, USA, 8-12 July 2006b. ACM Press. ISBN 1-59593-186-4. URL
http://www.cs.bham.ac.uk/~wbl/biblio/gecco2006/docs/p887.pdf.
GPBiB
E. P. K. Tsang, S. Markose, and H. Er. Chance discovery in stock index option and future
arbitrage. New Mathematics and Natural Computation, 1(3):435447, 2005.
E. P. K. Tsang and N. Jin. Incentive method to handle constraints in evolutionary.
In P. Collet, et al., editors, Proceedings of the 9th European Conference on Genetic
Programming, volume 3905 of Lecture Notes in Computer Science, pages 133144,
Budapest, Hungary, 10 - 12 April 2006. Springer. ISBN 3-540-33143-3. URL http:
//link.springer.de/link/service/series/0558/papers/3905/39050133.pdf. GPBiB
E. P. K. Tsang and J. Li. EDDIE for financial forecasting. In S.-H. Chen, editor, Genetic
Algorithms and Genetic Programming in Computational Finance, chapter 7, pages
161174. Kluwer Academic Press, 2002. ISBN 0-7923-7601-3. URL http://cswww.
essex.ac.uk/CSP/finance/papers/TsangLi-FGP-Chen_CompFinance.pdf.
GPBiB
E. P. K. Tsang, J. Li, and J. M. Butler. EDDIE beats the bookies. Software: Practice
and Experience, 28(10):10331043, 1998. ISSN 0038-0644. URL http://cswww.essex.
ac.uk/CSP/finance/papers/TsBuLi-Eddie-Software98.pdf.
GPBiB
E. P. K. Tsang, P. Yung, and J. Li. EDDIE-automation, a decision support tool for
financial forecasting. Decision Support Systems, 37(4):559565, 2004. URL http:
//cswww.essex.ac.uk/CSP/finance/papers/TsYuLi-Eddie-Dss2004.pdf.
GPBiB

BIBLIOGRAPHY

219

I. G. Tsoulos and I. E. Lagaris. Genanneal: Genetically modified simulated annealing.

Computer Physics Communications, 174(10):846851, 15 May 2006.
GPBiB
A. M. Turing. Intelligent machinery. Report for National Physical Laboratory. Reprinted
in Ince, D. C. (editor). 1992. Mechanical Intelligence: Collected Works of A. M. Turing.
Amsterdam: North Holland. Pages 107127. Also reprinted in Meltzer, B. and Michie,
D. (editors). 1969. Machine Intelligence 5. Edinburgh: Edinburgh University Press,
1948.
A. M. Turing. Computing machinery and intelligence. Mind, 49:433460, January 01
1950. URL http://www.cs.umbc.edu/471/papers/turing.pdf.
GPBiB
I. Usman, A. Khan, R. Chamlawi, and A. Majid. Image authenticity and perceptual
optimization via genetic algorithm and a dependence neighborhood. International
Journal of Applied Mathematics and Computer Sciences, 4(1):615620, 2007. ISSN
1305-5313. URL http://www.waset.org/ijamcs/v4/v4-1-7.pdf.
GPBiB
S. Vaidyanathan, D. I. Broadhurst, D. B. Kell, and R. Goodacre. Explanatory optimization of protein mass spectrometry via genetic search. Analytical Chemistry, 75(23):
66796686, 2003. URL http://dbkgroup.org/Papers/AnalChem75(6679-6686).pdf.
GPBiB
J. J. Valdes and A. J. Barton. Virtual reality visual data mining via neural networks
obtained from multi-objective evolutionary optimization: Application to geophysical
prospecting. In International Joint Conference on Neural Networks, IJCNN06, pages
48624869, Sheraton Vancouver Wall Centre Hotel, Vancouver, BC, Canada, 16-21
July 2006. IEEE.
GPBiB
V. Venkatraman, A. R. Dalby, and Z. R. Yang. Evaluation of mutual information and
genetic programming for feature selection in QSAR. Journal of Chemical Information
and Modeling, 44(5):16861692, 2004.
GPBiB
B. Vowk, A. S. Wait, and C. Schmidt. An evolutionary approach generates human competitive coreware programs. In M. Bedau, et al., editors, Workshop and Tutorial Proceedings Ninth International Conference on the Simulation and Synthesis of Living
Systems(Alife XI), pages 3336, Boston, Massachusetts, 12 September 2004. Artificial
Chemistry and its applications workshop.
GPBiB
I. Vukusic, S. N. Grellscheid, and T. Wiehe. Applying genetic programming to the prediction of alternative mRNA splice variants. Genomics, 89(4):471479, April 2007.
GPBiB
R.

L. Walker.
Search engine case study:
searching the web using genetic programming and MPI.
Parallel Computing, 27(1-2):7189, January
2001. URL http://www.sciencedirect.com/science/article/B6V12-42K5HNX-4/1/
57eb870c72fb7768bb7d824557444b72.
GPBiB

P. Walsh and C. Ryan. Paragen: A novel technique for the autoparallelisation of sequential programs using genetic programming. In J. R. Koza, et al., editors, Genetic Programming 1996: Proceedings of the First Annual Conference, pages 406
409, Stanford University, CA, USA, 2831 July 1996. MIT Press. URL http:
//cognet.mit.edu/library/books/view?isbn=0262611279.
GPBiB
D. C. Weaver. Applying data mining techniques to library design, lead generation and lead optimization. Current Opinion in Chemical Biology, 8(3):264270,
2004. URL http://www.sciencedirect.com/science/article/B6VRX-4CB69R1-2/2/
84a354cec9064ed07baab6a07998c942.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=71)

1.5

80
1
100
0.5

60
Fitness

Generation 71
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

220

BIBLIOGRAPHY

T. Weise and K. Geihs. Genetic programming techniques for sensor networks. In P. J.

Marron, editor, Proceedings of 5. GI/ITG KuVS Fachgesprach Drahtlose Sensornetze,
pages 2125, University of Stuttgart, July 2006. URL http://dgpf.sourceforge.net/
documents/006-2006-07-17-kuvs_paper.pdf. Technical Report No. 2006/07. GPBiB
C. H. Westerberg and J. Levine. Investigations of different seeding strategies in a
genetic planner. In E. J. W. Boers, et al., editors, Applications of Evolutionary Computing, volume 2037 of LNCS, pages 505514, Lake Como, Italy, 18 April
2001. Springer-Verlag. ISBN 3-540-41920-9. URL http://www.aiai.ed.ac.uk/~johnl/
papers/westerberg-evostim01.ps.
GPBiB
P. A. Whigham. A schema theorem for context-free grammars. In 1995 IEEE Conference on Evolutionary Computation, volume 1, pages 178181, Perth, Australia,
29 November - 1 December 1995. IEEE Press. URL http://citeseer.ist.psu.edu/
whigham95schema.html.
GPBiB
P. A. Whigham. Search bias, language bias, and genetic programming. In J. R. Koza,
et al., editors, Genetic Programming 1996: Proceedings of the First Annual Conference, pages 230237, Stanford University, CA, USA, 2831 July 1996. MIT Press.
GPBiB
L. D. Whitley. A Genetic Algorithm Tutorial. Statistics and Computing, 4:6585, 1994.
URL http://www.cs.colostate.edu/~genitor/MiscPubs/tutorial.pdf.
L. D. Whitley. An overview of evolutionary algorithms: practical issues and common
pitfalls. Information and Software Technology, 43(14):817831, 2001. URL http:
//www.cs.colostate.edu/~genitor/2001/overview.pdf.
GPBiB
L. D. Whitley, D. Goldberg, E. Cantu-Paz, L. Spector, I. Parmee, and H.-G. Beyer,
editors. Proceedings of the Genetic and Evolutionary Computation Conference
(GECCO-2000), Las Vegas, Nevada, USA, 10-12 July 2000. Morgan Kaufmann.
ISBN 1-55860-708-0. URL http://www.cs.colostate.edu/~genitor/GECCO-2000/
gecco2000mainpage.htm.
GPBiB
M. J. Willis, H. G. Hiden, and G. A. Montague. Developing inferential estimation algorithms using genetic programming. In IFAC/ADCHEM International Symposium on
Advanced Control of Chemical Processes, pages 219224, Banff, Canada, 1997a. GPBiB
M. Willis, H. Hiden, P. Marenbach, B. McKay, and G. A. Montague. Genetic programming: An introduction and survey of applications. In A. Zalzala, editor, Second International Conference on Genetic Algorithms in Engineering Systems: Innovations and Applications, GALESIA, University of Strathclyde, Glasgow, UK, 14 September 1997b. Institution of Electrical Engineers. ISBN 0-85296-693-8. URL
http://www.staff.ncl.ac.uk/d.p.searson/docs/galesia97surveyofGP.pdf. GPBiB
G. Wilson and M. Heywood. Introducing probabilistic adaptive mapping developmental
genetic programming with redundant mappings. Genetic Programming and Evolvable
Machines, 8(2):187220, June 2007. ISSN 1389-2576. Special issue on developmental
systems.
GPBiB
M. L. Wong and K. S. Leung. Evolving recursive functions for the even-parity problem
using genetic programming. In P. J. Angeline and K. E. Kinnear, Jr., editors, Advances
in Genetic Programming 2, chapter 11, pages 221240. MIT Press, Cambridge, MA,
USA, 1996. ISBN 0-262-01158-1.
GPBiB
M. L. Wong and K. S. Leung. Data Mining Using Grammar Based Genetic Programming
and Applications, volume 3 of Genetic Programming. Kluwer Academic Publishers,
January 2000. ISBN 0-7923-7746-X.
GPBiB

BIBLIOGRAPHY

221

M.-L. Wong, T.-T. Wong, and K.-L. Fok. Parallel evolutionary algorithms on graphics
processing unit. In D. Corne, et al., editors, Proceedings of the 2005 IEEE Congress on
Evolutionary Computation, volume 3, pages 22862293, Edinburgh, Scotland, UK, 2-5
September 2005. IEEE Press. ISBN 0-7803-9363-5. URL http://ieeexplore.ieee.
org/servlet/opac?punumber=10417&isvol=3.
A. M. Woodward, R. J. Gilbert, and D. B. Kell. Genetic programming as an analytical tool for non-linear dielectric spectroscopy. Bioelectrochemistry and Bioenergetics, 48(2):389396, 1999. URL http://www.sciencedirect.com/science/article/
B6TF7-3WJ72RJ-T/2/19fd01a6eb6ae0b8e12b2bb2218fb6e9.
GPBiB
S. Wright. The roles of mutation, inbreeding, crossbreeding and selection in evolution.
In D. F. Jones, editor, Proceedings of the Sixth International Congress on Genetics,
volume 1, pages 356366, 1932.
H. Xie, M. Zhang, and P. Andreae. Genetic programming for automatic stress detection
in spoken english. In F. Rothlauf, et al., editors, Applications of Evolutionary Computing, EvoWorkshops2006: EvoBIO, EvoCOMNET, EvoHOT, EvoIASP, EvoInteraction, EvoMUSART, EvoSTOC, volume 3907 of LNCS, pages 460471, Budapest, 1012 April 2006. Springer Verlag. ISBN 3-540-33237-5. URL http://www.springerlink.
com/openurl.asp?genre=article&issn=0302-9743&volume=3907&spage=460. GPBiB
L. Yamamoto and C. F. Tschudin. Experiments on the automatic evolution of protocols
using genetic programming. In I. Stavrakakis and M. Smirnov, editors, Autonomic
Communication, Second International IFIP Workshop, WAC 2005, Revised Selected
Papers, volume 3854 of Lecture Notes in Computer Science, pages 1328, Athens,
Greece, October 2-5 2005. Springer. ISBN 3-540-32992-7. URL http://cn.cs.unibas.
ch/people/ly/doc/wac2005-lyct.pdf.
GPBiB
D. Yamashiro, T. Yoshikawa, and T. Furuhashi. Visualization of search process and
improvement of search performance in multi-objective genetic algorithm. In G. G. Yen,
et al., editors, Proceedings of the 2006 IEEE Congress on Evolutionary Computation,
pages 11511156, Vancouver, BC, Canada, 16-21 July 2006. IEEE Press. ISBN 0-78039487-9. URL http://ieeexplore.ieee.org/servlet/opac?punumber=11108.
K. Yanai and H. Iba. Estimation of distribution programming based on bayesian network.
In R. Sarker, et al., editors, Proceedings of the 2003 Congress on Evolutionary Computation CEC2003, pages 16181625, Canberra, 8-12 December 2003. IEEE Press. ISBN
0-7803-7804-0. URL http://www.iba.k.u-tokyo.ac.jp/papers/2003/yanaiCEC2003.
pdf.
GPBiB
K. Yanai and H. Iba. Program evolution by integrating EDP and GP. In K. Deb, et al.,
editors, Genetic and Evolutionary Computation GECCO-2004, Part I, volume 3102
of Lecture Notes in Computer Science, pages 774785, Seattle, WA, USA, 26-30 June
2004. Springer-Verlag. ISBN 3-540-22344-4. URL http://www.iba.k.u-tokyo.ac.jp/
papers/2004/yanaiGECCO2004.pdf.
GPBiB
M. Yangiya. Efficient genetic programming based on binary decision diagrams. In 1995
IEEE Conference on Evolutionary Computation, volume 1, pages 234239, Perth, Australia, 29 November - 1 December 1995. IEEE Press.
GPBiB
X. Yao, E. Burke, J. A. Lozano, J. Smith, J. J. Merelo-Guerv
os, J. A. Bullinaria, J. Rowe,
P. T. A. Kab
an, and H.-P. Schwefel, editors. Parallel Problem Solving from Nature - PPSN VIII, volume 3242 of LNCS, Birmingham, UK, 18-22 September 2004.
Springer-Verlag. ISBN 3-540-23092-0. URL http://www.springerlink.com/openurl.
asp?genre=issue&issn=0302-9743&volume=3242.
GPBiB
100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=78)

1.5

80
1
100
0.5

60
Fitness

Generation 78
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

222

BIBLIOGRAPHY

Yoshihara. Proceedings of the 2000 Congress on Evolutionary Computation CEC00, La

Jolla Marriott Hotel La Jolla, California, USA, 6-9 July 2000. IEEE Press. ISBN
0-7803-6375-2.
GPBiB
J. Yu and B. Bhanu. Evolutionary feature synthesis for facial expression recognition.
Pattern Recognition Letters, 27(11):12891298, August 2006. Evolutionary Computer
Vision and Image Understanding.
GPBiB
J. Yu, J. Yu, A. A. Almal, S. M. Dhanasekaran, D. Ghosh, W. P. Worzel, and A. M.
Chinnaiyan. Feature selection and molecular classification of cancer using genetic
programming. Neoplasia, 9(4):292303, April 2007.
GPBiB
T. Yu. Hierachical processing for evolving recursive and modular programs using higher
order functions and lambda abstractions. Genetic Programming and Evolvable Machines, 2(4):345380, December 2001. ISSN 1389-2576.
GPBiB
T. Yu and S.-H. Chen. Using genetic programming with lambda abstraction to find technical trading rules. In Computing in Economics and Finance, University of Amsterdam,
8-10 July 2004.
GPBiB
T. Yu, R. L. Riolo, and B. Worzel, editors. Genetic Programming Theory and Practice
III, volume 9 of Genetic Programming, Ann Arbor, 12-14 May 2005. Springer. ISBN
0-387-28110-X.
GPBiB
B.-T. Zhang and D.-Y. Cho. Coevolutionary fitness switching: Learning complex collective behaviors using genetic programming. In L. Spector, et al., editors, Advances
in Genetic Programming 3, chapter 18, pages 425445. MIT Press, Cambridge, MA,
USA, June 1999. ISBN 0-262-19423-6. URL http://bi.snu.ac.kr/Publications/
Books/aigp3.ps.
GPBiB
B.-T. Zhang and H. M
uhlenbein. Evolving optimal neural networks using genetic algorithms with Occams razor. Complex Systems, 7:199220, 1993. URL http:
//citeseer.ist.psu.edu/zhang93evolving.html.
GPBiB
B.-T. Zhang and H. M
uhlenbein. Balancing accuracy and parsimony in genetic programming. Evolutionary Computation, 3(1):1738, 1995. URL http://www.ais.
fraunhofer.de/~muehlen/publications/gmd_as_ga-94_09.ps.
GPBiB
B.-T. Zhang, P. Ohm, and H. M
uhlenbein. Evolutionary induction of sparse neural
trees. Evolutionary Computation, 5(2):213236, 1997. URL http://bi.snu.ac.kr/
Publications/Journals/International/EC5-2.ps.
GPBiB
M. Zhang and U. Bhowan. Pixel statistics and program size in genetic programming for
object detection. Technical Report CS-TR-04-3, Computer Science, Victoria University
of Wellington, New Zealand, 2004.
GPBiB
M. Zhang and W. Smart. Using gaussian distribution to construct fitness functions in genetic programming for multiclass object classification. Pattern Recognition Letters, 27
(11):12661274, August 2006. Evolutionary Computer Vision and Image Understanding.
GPBiB
Q. Zhang, J. Sun, and E. P. K. Tsang. An Evolutionary Algorithm With Guided Mutation
for the Maximum Clique Problem. IEEE Transactions on Evolutionary Computation,
9(2):192200, 2005.

BIBLIOGRAPHY

223

Y. Zhang and P. I. Rockett. Evolving optimal feature extraction using multi-objective

genetic programming: a methodology and preliminary study on edge detection. In H.G. Beyer, et al., editors, GECCO 2005: Proceedings of the 2005 conference on Genetic
and evolutionary computation, volume 1, pages 795802, Washington DC, USA, 25-29
June 2005. ACM Press. ISBN 1-59593-010-8. URL http://www.cs.bham.ac.uk/~wbl/
biblio/gecco2005/docs/p795.pdf.
GPBiB
Y. Zhang and P. I. Rockett. Feature extraction using multi-objective genetic programming. In Y. Jin, editor, Multi-Objective Machine Learning, volume 16 of Studies in
Computational Intelligence, chapter 4, pages 79106. Springer, 2006. ISBN 3-54030676-5. Invited chapter.
GPBiB
E. Zitzler, M. Laumanns, and L. Thiele. SPEA2: Improving the Strength Pareto Evolutionary Algorithm. Technical Report 103, Gloriastrasse 35, CH-8092 Zurich, Switzerland, 2001. URL http://citeseer.ist.psu.edu/article/zitzler01spea.html.

100

1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=85)

1.5

80
1
100
0.5

60
Fitness

Generation 85
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

Index
% protected division, 22
22 bit parity, sub-machine-code GP, 93
abstraction operator, 48
active code, 102
adaptive market hypothesis, 123
ADATE, 48
ADF
crossover, 49
recursion prevention, 49
troubleshooting, 134
agent
evolutionary, 122
image processing, 122
social simulation, 123
aggregate fitness function, 7576
analogue circuit evolution, 119
ant colony optimisation (ACO), 74
PEEL, 74
ant programming, generalised (GAP), 74
ant-TAG, 74
anytime fitness, 85
applications
arbitrage, 123
architecture, 76
art, 127
bin packing, 127
bioinformatics, 125126
biology, 125
bomb disposal, 124
cellular automata, 119
chemical engineering, 124
chemistry, 126
circuits, 119
classification, 121
control, 119
curve fitting, 113
data compression, 128
data mining, 85, 125
data modelling, 113
economics, 123

entertainment, 127
exchange rate, 123
film industry, 128
finance, 123
gambling, 123
games, 120, 127
guidelines, 111
heuristics, 126
human competitive, 117
hyper-heuristics, 126
image processing, 121, 122
industry, 124
infrared spectra, 125
jet engine optimisation, 125
mechanics, 120
medical, 122, 125
meta-heuristics, 126
music, 128
neural networks, 121
nuclear reactor control, 124
numerical control, 124
OCR, 122
patents, 119
population size, 116
process control, 124
QSAR, 78
quantum computing, 118
robotics, 118, 119
SAT, 127
side effects, 113
signal processing, 121
stock market, 123
symbolic regression, 113
teaching aids, 124
TSP, 127
video, 129
watermarking, 122
arbitrage, 123
architecture
evolution, 50
program, 50

225

226
architecture altering operator, 48, 50
architecture design tool, 76
architecture-defining preparatory step, 50
arity
function, 11
node, 11
art, evolutionary, 127128
artificial intelligence, 1
human competitive, 117121
artificial neural network, 66, 121
auto parallelising code, 68
automatically defined function (ADF), 48
limitations, 49
primitive sets, 50
automatically defined iteration (ADI), 50
automatically defined loop (ADL), 50
automatically defined recursion (ADR), 50
automatically defined store (ADS), 50
Backgammon, 120, 127
backward chaining GP, 86
Bayesian automatic programming (BAP),
74
Bayesian optimisation algorithm (BOA), 70
Beagle, Open, C++ implementation, 148
beowulf cluster, 95
bias induced by constraints, 55
bibliography, GP, 148
bin packing problem, 127
bioinformatics, 125, 126
biological system, 125
bloat, 97, 101
control, 76
crossover bias theory, 104
definition, 101
executable model, 102
initialisation effects, 40
lexicographic control, 77
lexicographic parsimony, 77, 78, 80
MDL, 107
multi-objective, 77
mutation effects, 42
nature of program search spaces theory, 102
Pareto control, 78
parsimony pressure, 106
practical effects, 101
removal bias theory, 102
replication accuracy theory, 102
size evolution equation, 103
Tarpeian method, 106
vs. growth, 101
boids, 128
bomb disposal, 124
books, GP, 146

INDEX
branching factor, 15
bulletin board, 149
cache
data, 86
Cartesian GP, 67
GPU, 90
multi-objective, 81
CCNOT, distribution of circuits, 99
cellular automaton, 68
evolution, 119
cellular encoding, 57
check pointing, 85
Checkers, 127
chemical engineering, 124
chemistry, 78, 126
Chess, 120, 127
circuit design
multi-objective, 76
circuit, analogue
evolution, 119
classification, 132
cellular automaton, majority vote, 68
handwriting, 122
infrared images, 121
M25, 121
optical, 122
PADO, 67
SAR radar, 121
sonar, 122
closure, 2122
cluster computing, 95
co-evolution, 46
code
active, 102
inactive, 102
communication topology
ring, 94
toroidal grid, 94
compiled vs. interpreted GP, 24
GPU, 90
compiling GP populations, 90
comprehensible programs, 51
compression
fractal, 129
image, 128
lossless, 129
lossy, 128, 129
sound, 128
video, 129
wavelet, 129
computational chemistry, 126
computer art, 127128
computer game, 127
computer program, evolution, 2

INDEX

227

conference submission, 139

conferences, 145, 147
constant, ephemeral random, 20
constraints
bias, 55
enforced by crossover etc, 53
grammar-based, 53
on tree size or depth, 104
dynamic, 105
on tree structure, 52
reduction of search space, 55
semantic, 56
strong typing, 52
syntactic, 56
context-preserving crossover, 45
theory, 98
context-sensitive grammar, 55
control
proportional integrative and derivative (PID), 119
robot, 119
Core War, 120
covariant parsimony pressure, 107
crossover
between ADFs, 49
bias theory, 104
Cartesian GP, 68
constrained, 52
context-preserving, 45, 98
depth based, 45
example, 34
FPGA, 93
headless chicken, 16
homologous, 44, 98
linear GP, homologous, 64
one-point, 44
operator, 2
PDGP, 66
point, 15
rate, 17
size-fair, 45, 98, 105
subtree, 15
uniform, 44
curve fitting, 113116
data cache, 86
data flow GP, 65
data mining, 85
many variables, 125
medical, 125
data modelling, 113116
data visualisation, 135
virtual reality, 80
deme, 93
ring topology, 94
100

toroidal grid, 94
density of solutions, 57
depth
node, 12
tree, 12
depth limits, 104
derivation tree, 54
developmental GP, 23, 57
direct problem, 112
directed acyclic graph (DAG), 87
DirectX, 90
disassortative mating, 82
Discipulus, 63, 124
discussion group, 149
distributed
evolutionary algorithm, 8895
GP, 9395
populations
fine-grained, 94
geographically, 9395
ring topology, 94
toroidal grid, 94
distribution
sampling, 69
diversity, 94
promotion, multi-objective GP, 78
Draughts, 127
dynamic fitness, 84
dynamic size or depth limits, 105
dynamic subset selection (DSS), 85
EA, 1
EC, 1
ECJ, Java implementation, 148
economic modelling, 123124
editing operator, 46
efficient market hypothesis, 123
elitism, troubleshooting, 137
Elvis robot, 115
embarrassingly parallel, 89
engine monitoring and control, 121
entropy, 129
ephemeral random constant, 20
example, 29
estimation of distribution algorithms
(EDAs), 69
estimation of distribution programming
(EDP), 72
evaluation safety, 22
even parity, 78
evolutionary algorithm (EA), 1
distributed, 8895
evolutionary art, 128
evolutionary computation (EC), 1
evolutionary music, 128
1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=92)

1.5

80
1
100
0.5

60
Fitness

Generation 92
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

228
evolutionary quantum computing, 120
evolutionary search, mathematical models,
98
evolving agents, 122
evolving designs, 23
example
crossover, 34
ephemeral random constant, 29
fitness, 30
function set, 30
mutation, 33
parameter, 30
terminal set, 29
termination, 31
exceptions
integer overflow, 22
problems caused by trapping, 22
exchange rate, 123
executable model, bloat, 102
expression simplification, 135
extended compact genetic algorithm
(eCGA), 70
extended compact GP (eCGP), 72
farming in parallel GP, 89
feature selection, 115
field programmable gate array (FPGA), 92
films, feature, 128
films, GP, 146
financial time series prediction, 123124
fine-grained distributed GP, 94
finite state automata, 66
fitness, 2, 2426
anytime, 85
case, 26
reduction, 83
combined objectives, 7576
dynamic, 84
dynamic, multi-objective, 8081
example, 30
fast
FPGA, 92
GPU, 9092
sub-machine-code, 93
hits, 76
incremental, 58
multi-objective, 7580
image processing, 122
RMS, 115
sharing, 77, 137
staged, 81, 85
static, problems with, 84
symbolic regression, 115
for, syntax constraint, 52
Fourier transform, quantum evolution, 120

INDEX
FPGA, 92
GP implementation, 93
fractal compression, 129
freeware, 148
ECJ, 148
GPC++, 148
Lil-GP, 148
Open Beagle, 148
TinyGP, 151162
frequency of primitives, 136
full random trees, 1213
function arity, 11
function set, 1923
evolving non-programs, 23
example, 30
modelling, 115
side effects, 24
sufficiency, 2223
function-defining branch, 48
gambling, 123
game theory, 123
games, 127
generalisation-accuracy tradeoff, 107
generalised ant programming (GAP), 74
genetic operator, 2
rates, 26, 116
genetic program representation, 9
GENR8, 76
geographically distributed GP, 9395
GP implementations, 147148
TinyGP, 151162
GP problem solver (GPPS), 51
GP-ZIP, 130
GPC++, implmentation, 148
GPU, 90
speedup factor, 92
grammar, 53
based constraint, 53
based GP, 53
initialisation, 57
operators, 57
context-sensitive, 55
tree adjoining, 55
grammar model based program evolution
(GMPE), 74
grammatical evolution
troubleshooting, 134
grammatical evolution, 55
graph, directed acyclic (DAG), 87
graph-based GP, 6568
graphics processing unit (GPU), 90
grow random trees, 1214
tree size bias, 13
growth vs. bloat, 101

INDEX

229

guidelines, applications, 111

halting probability, 100
halting program, Markov chain model, 100
Hamming distance, 71
hardware implementations of GP, 93
headless chicken crossover, 16
heuristic, 126
hierarchical structure, 47
higher-order type systems, 53
hill climbing, 46
hits, 76
hoist mutation, 43, 106
homologous crossover, 44
linear GP, 64
theory, 98
Huffman encoding, 129
human competitive
artificial intelligence, 117121, 141
awards (Humies), 119
human readable programs, 51
hyper-heuristic, evolving, 126
image
compression, 128
processing, 121, 122
watermark, 122
implementations of GP, 147148
FPGA, 93
GPU, 9092
TinyGP, 151162
inactive code, 102
incremental fitness, 58
indirect representation, 58
industrial modelling, multi-objective GP,
79
information content of population, 139
information theory, 46
infrared images, 121
infrared spectra, 125
initialisation
effects on bloat, 40
full method, 1213
grammar-based GP, 57
grow method, 1214
tree size bias, 13
Lagrange, 41
ramped half-and-half, 13
ramped uniform, 40
type system, 56
uniform, 40
integer overflow, 22
intelligence, artificial (AI), 1
interpreted GP, 24
linear, 62, 64
100

Java, 64
T7, 64
SIMD and GPU, 92
introns
troubleshooting, 136
useful with mutation, 64
invention, evolution of, 119
inverse kinematics, 115
inverse problem, 112
iterated functions system (IFS), 128
jet engine optimisation, 125
journals, 147
kinematics, 120
L-system, 74, 76
tree adjoining grammar, 58
Lagrange
distribution of the second kind, 104
initialisation, 41
large populations, 137
learning, machine (ML), 1
lexicographic
parsimony bloat control, 77, 80
preference, 77, 80
libraries, dynamic, 47, 48
Lil-GP, 148
limits, size and depth, 104
Lindenmayer grammar, see L-system
linear GP, 6164
Cartesian GP, 67
crossover, 64
homologous, 64
instruction format, 62, 63
interpreted, 62, 64
introns, 64
removal, 64
Java, 64
mutation, 64
speed, 62
T7, 64
linear representation, 6164
Cartesian GP, 67
linearised tree-based GP
prefix, 62
reverse polish, 92
logic network, evolution, 66
lossless compression, 129
lossy compression, 128
machine code GP, 6264
Intel x86, 63
SPARC, 62, 63
Z80, 62, 63
1000

Avg Size

Avg Fitness
Best Fitness

90

sin(x)
GP (gen=99)

1.5

80
1
100
0.5

60
Fitness

Generation 99
(see Sec. B.4)

Average Size

70

50

40

-0.5
10

30
-1
20
-1.5

10

20

40

60
Generations

80

100

20

40

60
Generations

80

100

-2

4
x

230
machine intelligence, 1
human-competitive, 141
machine learning (ML), 1
mailing list, GP, iii, 149
Markov chain model
evolutionary algorithms, 98
GP, 98
program execution, 100
masterslave GP, 89
max problem, 72
medical imaging, 122
message passing interface (MPI), 95
meta-heuristics, 126
meta-optimising semantic evolutionary
search (Moses), 72
migration rate, 94
MIMIC, 70
minimum description length (MDL), 46,
107
model car racing, 127
model, executable, 102
modular structure, 47
modules, 59
Moores Law, 90
MPI, 95
multi-level type systems, 53
multi-objective fitness, 7576
image processing, 122
multi-objective GP (MO GP), 41, 75
jets, 125
operator pressure, 81
Pareto dominance, 7680
small vs good, 81
multi-objective optimisation (MOO), 75
data visualisation, 80
preference information, 77
multiple typed programs, 2122
music, evolutionary, 128
mutation, 2
constants, 43
dynamic libraries, 48
example, 33
hoist, 43, 106
local search, 43
node replacement, 43
permutation, 43
point, 16, 43
rate, 17
shrink, 43, 106
simulated annealing, 44
size-fair, 42, 105
subtree, 16
survey, 4244
swap, 43

INDEX
N-gram GP, 72
NAND, distribution of circuits, 99
nature of program search spaces theory, 102
neural network evolution, 121
with PDGP, 66
niching, 77
NLP
parsing and tagging, multi-objective
GP, 79
text retrieval, multi-objective GP, 79
node
arity, 11
depth, 12
replacement mutation, 43
non-terminal symbols, 53
non-Turing complete program, theory, 99
NSGA-II, extension to GP, 79
nuclear reactor control, 124
numeric regression, 30
numerical control, 124
Odin, 51
one-max, 72
one-point crossover, 44
theory, 98
OpenGL, 92
operator
architecture-altering, 50
composition, 17
constrained, 53
crossover, 2, 34
editing, 46
genetic, 2
grammar-based GP, 57
mutation, 2
rate, 17
reorganisation of subtrees, 46
repair, 81
reproduction, 17, 33
optical character recognition (OCR), 122
Oscar, 128
Othello, 127
over-fitting, 46, 139, 140
dynamic fitness function, 84
overflow, numeric, 22
Pac-Man, Ms, 127
PADO
anytime programming, 67
random access memory, 67
panmictic population, 137
Paragen, 68
parallel computing, 88
parallel distributed GP (PDGP), 65
ADF, 66

INDEX
link set, 66
no side effects, 67
parallel evolutionary algorithm, 8895
parallel GP, 89, 94
parallel processing, 90
parallel virtual machine (PVM), 95
parameter
quadratic polynomial example, 30
values, 2627, 116
Pareto
criterion, 106
dominance, 7680
relaxation, 78
front, 77
preference, 80
tournament, 78
Parisian GP, 122
parsimony
coefficient, 106
difficulties, 107
dynamic, 106, 107
lexicographic, 77
pressure, 106
covariant, 107
particle swarm optimisation (PSO), 127
patentable invention, 119
penalty bloat control method, 107
permutation mutation, 43
point mutation, 16, 43
Poker, 127
polymorphic types, 53
population size, 2627
applications, 116
troubleshooting, 134
population variety, 136
population-based incremental learning
(PBIL), 70
post fix notation, 92
precision vs. recall, multi-objective GP, 79
premature convergence, 84
preparatory step
architecture-defining, 50
fitness function, 24
function set, 20
parameters, 26
terminal set, 19
termination, 27
primitive set, 9, 1923
evolving non-programs, 23
modelling, 115
side effects, 20, 24
vector-based, 51
primitives, number in population, 136
probabilistic incremental program evolution (PIPE), 71

231
probability distribution
sampling, 69
process control, 124125
production rule, 53
program
architecture, 50
evolution, 50
evolution, 2
human understandability, 51
program evolution with explicit learning
(PEEL), 74
program execution, 24
programmatic compression, 128, 129
proportional integrative and derivative
(PID), controller, 119
proposals to funding agencies, 139
protected
MOVE AHEAD, 22
division, 22
other operations, 22
pruning neural networks, 82
PushGP, 59
PVM, 95
pygmies and civil servants, 81
QSAR, multi-objective, 78
quantum
algorithm evolution, 118
computing, 118, 120
Fourier transform evolution, 120
quantum programs, distribution of, 99
ramped half-and-half, 11, 13
problems with, 40
ramped uniform initialisation, 40
RapidMind, 92
rate
crossover, 17
mutation, 17
operator, 17
rational allocation of trials, 84
recurrent transition network, evolution, 66
recursion, 59
regression, 114
numeric, 30
symbolic, 30
applications, 113116
removal bias theory, 102
reorganisation of subtrees, 46
repair operator, 81
replication accuracy theory, 102
representation
GP, 9
graph-based, 6568
prefix notation, 10

232
reverse polish, 92
syntax tree, 9
tree-based in GP, 10
reproduction operator, 17
in example, 33
result-producing branch, 48
reusable component, 48
reverse polish representation, 92
reversible computing, distribution of, 99
rewrite rule, 53
ring species, 88
RoboCup, 118
robot
control, 119
Elvis, 115
football, 127
run time errors, avoiding, 22
run time exceptions, problems, 22
sampling
probability distribution, 69
SAT, 127
schema theory, 98
search
evolutionary models, 98
stochastic, 97
search space, 24, 97, 99
seeding, 41
selection of parents
tournament, 1415
backward chaining GP, 86
semantic constraints, 56
sensor, soft, 114
seti.home, 95
shrink mutation, 43, 106
side effects, 20, 24
applications, 113
signal processing, 121122
SIMD, 90
simple generics, 53
simulated annealing, 44, 46
single typed programs, 21, 51
size
evolution equation, 103
Lagrange distribution, 104
limits, 104
tree, 12
size-fair
crossover, 45, 105
theory, 98
mutation, 42, 105
soft sensor, 114
solution density, 57
sound compression, 128
spatial separation, 94

INDEX
SPEA2, 78
species
formation, 88
ring, 88
speedup factor, GPU, 92
speedup techniques, 83
staged fitness function, 81
static fitness, problems with, 84
stochastic computing, distribution of, 99
stochastic context-free grammar, 74
stochastic search, 97
stock market, 123
Stroganoff, 43
strongly typed GP, 52
theory, 98
sub-machine-code GP (SMCGP), 93
sub-population, 93
submitting to a journal, 139
subtree
crossover, 15
mutation, 16
reorganisation, 46
sufficiency, 2223
supercomputer, 89, 95
swap mutation, 43
swarm, 122
symbolic regression, 30
applications, 113116
syntactic constraints, 56
syntax tree, 9
T7 programming language, 64, 100
take over time, 84
Tarpeian bloat control, 106
teaching aids, 124
terminal set, 1920
evolving non-programs, 23
example, 29
modelling, 115
side effects, 20, 24
sufficiency, 2223
terminals of a grammar, 53
termination
criterion, 27
example, 31
theory, 97
bloat, 104
crossover
bias, 104
context-preserving, 98
homologous, 98
one-point, 98
size-fair, 98
non-Turing complete program, 99
schema, 98

INDEX
search spaces, 99
thesis write up, 139
time series, financial, 123124
TinyGP, 151162
Toffoli, distribution of circuits, 99
tournament selection, 1415, 86
Transputer, 94
trap function, 72
travelling salesman problem (TSP), 127
tree
depth, 12
derivation, 54
editing, 46
size, 12
tree adjoining grammar (TAG), 55
tree-based representation, 10
Tron, 127
Turing complete GP, 64
Turing complete program, theory, 99
Turing test, 117, 142
type
consistency, 2122, 51
conversion, 51
initialisation, 56
more understandable programs, 51
multiple, 52
single, 51
strong, 52
system
higher-order, 53
multi-level, 53
unexploded bombs, 124
uniform crossover, 44
uniform initialisation, 40
uniform multivariate distribution algorithm
(UMDA), 70
validating results, 132
variety, of population, 136
vector-based primitive, 51
VHDL, multi-objective GP, 79
video compression, 129
videos, GP, 146
virtual reality data visualisation, 80
watermark security
application, 122
multi-objective GP, 79
wavelet lossy compression, 129
world wide GP, 95
Wrights geographic model of evolution, 88

233

Colophon
This book was primarily written using the LATEX document preparation
system, along with BibTEX, pdflatex and makeindex. Most of the editing
was done using the emacs and xemacs editors, along with extensions such
as RefTEX; some was done with TEXShop as well. Most of the data plots
were generated using gnuplot and the R statistics package. Diagrams were
generated with a variety of tools, including the Graphviz package, tgif and
xfig. A whole host of programming and scripting languages were used to
automate various processes in both the initial scientific research and in the
production of this book; they are too numerous to list here, but were crucial
nonetheless. The cover was created with Adobe Photoshop1 and gimp.
Coordinating the work of three busy, opinionated authors is not trivial,
and would have been much more difficult without the use of revision control
systems such as Subversion. Around 500 commits were made in a six month
period, averaging around 10 commits per day in the final weeks. The actual
files were hosted as a project at http://assembla.com; we didnt realise
until several months into the project that Assemblas president is in fact
Andy Singleton, who did some cool early work in GP in the mid-90s.
The reviews and summaries on the back cover were generated
stochastically using the idea of N-grams from linguistics. For the reviews
we collected a number of reviews of previous books on GP and EAs, and
tabulated the frequency of different triples of adjacent words. These frequencies of triples in the source text were then used to guide the choices of
words in the generated reviews. The only word following the pair ad
and hoc in our source reviews, for example, was tweaks; thus once ad
and hoc had been chosen, the next word had to be tweaks. The pair
of the, on the other hand, appears numerous times in our source text, followed by words such as field, body, and rapidly. However, theory
is the most common successor, and, therefore, the most likely to be chosen to follow of the in the generation of new text. The generation of the
summaries was similar, but based on the front matter of the book itself.
See (Poli and McPhee, 2008a) for an application of these ideas in genetic
programming.
1 Adobe

Photoshop is a registered trademark of Adobe Systems Incorporated