Genetic Programming

Krzysztof Krawiec
Tomasz Pawlak

Laboratory of Intelligent Decision Support Systems

Institute of Computing Science, Poznan University of Technology, Pozna«, Poland
http://www.cs.put.poznan.pl/kkrawiec/

1
 Introduction

Introduction 2
 Outline and objectives

1 Introduction: GP as a variant of EC
2 Specic features of GP
3 Variants of GP
4 Applications
5 Some theory
6 Case studies

Introduction 3
 Bilbiography

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

of Natural Selection MIT Press, 1992
A Field Guide to Genetic Programming (ISBN 978-1-4092-0073-4)
http://www.gp-eld-guide.org.uk/
Langdon, W. B. Genetic Programming and Data Structures: Genetic
Programming + Data Structures = Automatic Programming! Kluwer, 1998
Langdon, W. B. & Poli, R. Foundations of Genetic Programming Springer-Verlag,
2002
Riolo, R. L.; Soule, T. & Worzel, B. (ed.) Genetic Programming Theory and
Practice V Springer, 2007
Riolo, R.; McConaghy, T. & Vladislavleva, E. (ed.) Genetic Programming Theory
and Practice VIII Springer, 2010
See: http://www.cs.bham.ac.uk/~wbl/biblio/

Introduction 4
 Acknowledgment

Credits go to:
A Field Guide to Genetic Programming http://www.gp-field-guide.org.uk/
[42]

Introduction 5
 Background

Background 6
 Evolutionary Computation (EC)

Heuristic bio-inspired global search algorithms

Operate on populations of candidate solutions
Candidate solutions are encoded as genotypes
Genotypes get decoded into phenotypes when evaluated by the tness function f
being optimized.

Formulation:

p ∗ = arg max f (p )
p∈P
where
P is the considered space (search space) of candidate solutions (solutions for
short)
f is a (maximized) tness function
p ∗ is an optimal solution (an ideal) that maximizes f .

Background 7
 Generic evolutionary algorithm

Evolutionary Algorithm

Initialization of population P

Population P of individuals

Solution/individual s
Evaluation Fitness function f
f(s)
Termination criteria

Selection

Mutation and recombination

Output: Best solution s+

Background 8
 [Unique] characteristic of EC

Black-box optimization (f 0 s dependency on the independent variables does not

have to be known or meet any criteria)
Fining an optimum cannot be guaranteed, but in practice a well-performing
suboptimal solution is often satisfactory.
Importance of crossover: a recombination operator that makes the solutions
exchange certain elements (variable values, features)
Without crossover: parallel stochastic local search

Background 9
 What is genetic programming?

What is genetic programming? 10

 Genetic programming

In a nutshell:
A variant of EC where the genotypes represent programs, i.e., entities capable of
reading in input data and producing some output data in response to that input.
Fitness function f measures the similarity of the output produced by the program
to the desired output, given as a part of task statement.
Standard representation: expression trees.

Important implication: Additional input required by the algorithm (compared to EC):

Set of instructions (programming language of consideration).
Data to run the programs on.

What is genetic programming? 11

 Conceptual preliminaries of GP: Space of candidate solutions P

Candidate solutions p ∈ P evolving under the selection pressure of the tness

function f are themselves functions of the form p : I → O ,
I and O are, respectively, the spaces of input data and output data accepted and
produced by programs from P .
Cardinality of |P | is typically large or innite.

The set of program inputs I , even if nite, is usually so large that running each
candidate solution on all possible inputs becomes intractable.
GP algorithms typically evaluate solutions on a sample I 0 ⊂ I , |I 0 |  |I | of possible
inputs, and tness is only an approximate estimate of solution quality.
The task is given as a set of tness cases, i.e., pairs (xi , yi ) ∈ I × O , where xi
usually comprises one or more independent variables and yi is the output variable.

What is genetic programming? 12

 Conceptual preliminaries: Fitness function f

In most cases (and most real-world applications of GP), tness function f

measures the similarity of the output produced by the program to the desired
output, given as a part of task statement.
Then, tness can be expressed as a monotonous function of the divergence of
program's output from the desired one, for instance as:

f (p ) = − ∑ ||yi − p (xi )||, (1)

i
where
p (xi ) is the output produced by program p for the input data xi ,
|| · || is a metric in the output space O ,
i iterates over all tness cases.

What is genetic programming? 13

 Conceptual preliminaries: Character of candidate solutions

The candidate solutions in GP are being assembled from elementary entities called
instructions.
A part of formulation of a GP task is then also an instruction set I , i.e., a set of
symbols used by the search algorithm to compose the programs (candidate
solutions).
Design of I usually requires some background knowledge;
In particular, it should comprise all instructions necessary to nd solution to the
problem posed (closure).

What is genetic programming? 14

 Genetic programming

Main evolution loop (`vanilla GP')

1: procedure GeneticProgramming(f , I ) . f - tness function, I - instruction set

2: P ← {p ← RandomProgram(I )} . Initialize population
3: repeat . Main loop over generations
4: for p ∈ P do . Evaluation
5: p .f ← f (p ) . p .f is a `eld' in program p that stores its tness
6: end for
7: 0
P ← 0/ . Next population
8: repeat . Breeding loop
9: p1 ← TournamentSelection(P) . First parent
10: p2 ← TournamentSelection(P) . Second parent
11: (o1 , o2 ) ← Crossover(p1 , p2 )
12: o1 ← Mutation(o1 , I )
13: o2 ← Mutation(o2 , I )
14: P 0 ← P 0 ∪ {o1 , o2 }
15: until |P 0 | = |P|
16: P ← P0
17: until StoppingCondition(P )
18: return arg maxp∈P p .f
19: end procedure

What is genetic programming? 15

 Crossover

Crossover: exchange of randomly selected subexpressions (subtree swapping crossover).

1: function Crossover(p1 , p2 )
2: repeat
3: s1 ← Random node in p1
4: s2 ← Random node in p2
5: (p10 , p20 ) ← Swap subtrees rooted in s1 and s2
6: until Depth(p10 ) < dmax ∧ Depth(p20 ) < dmax . dmax is the tree depth
limit
7: return (p10 , p20 )
8: end function

What is genetic programming? 16

 Mutation

Mutation: replace a randomly selected subexpression with a new randomly generated

subexpression.

1: function Mutation(p , I )
2: repeat
3: s ← Random node in p
4: s 0 ← RandomProgram(I )
5: p 0 ← Replace the subtree rooted in s with s 0
6: until Depth(p 0 ) < dmax . dmax is the tree depth limit
7: return p 0
8: end function

What is genetic programming? 17

 Exemplary run: Setup

Objective: Find program whose output matches x 2 + x + 1 over the range [−1, 1].
Such tasks can be considered as a form of regression.
As solutions are built by manipulating code (instructions), this is referred to as
symbolic regression.

Fitness: sum of absolute errors for x ∈ −1.0, −0.9, ...0.9, 1.0

In other words, the set of tness cases is:
xi -1.0 -0.9 ... 0 ... 0.9 1.0
yi 1 0.91 ... 1 ... 2.71 3

What is genetic programming? 18

 Exemplary run: Setup

Instruction set:
Nonterminal (function) set: +, -, % (protected division), and ∗; all operating on
oats
Terminal set: x , and constants chosen randomly between -5 and +5
Selection: tness proportionate (roulette wheel) non elitist
Initial pop: ramped half-and-half (depth 1 to 2. 50% of terminals are constants)
(to be explained later)
Parameters:
population size 4,
50% subtree crossover,
25% reproduction,
25% subtree mutation, no tree size limits
Termination: when an individual with tness better than 0.1 found

What is genetic programming? 19

 Initial population (population 0)

What is genetic programming? 20

 Fitness assignment for population 0

Fitness values: f(a)=7.7, f(b)=11.0, f(c)=17.98, f(d)=28.7

What is genetic programming? 21
 Breeding

Assume:
a gets reproduced
c gets mutated (at loci 2)
a and d get crossed-over
a and b get crossed over

What is genetic programming? 22

 Population 1
Population 0:

Population 1:

Individual d in population 1 has tness 0.

What is genetic programming? 23
 Summary of our rst glimpse at GP

Summary of our rst glimpse at GP 24

 Specic features of GP

The solutions evolving under the selection pressure of the tness function are
themselves functions (programs).
GP operates on symbolic structures of varying lengths.
There are no variables for the algorithm to operate on (at least in the common
sense).
The program can be tested only on a limited number of tness cases (tests).

=⇒ In contrast to most EC methods that are typically placed in optimization

framework, GP is by nature an inductive learning approach that ts into the domain of
machine learning [33].

Summary of our rst glimpse at GP 25

 In a broader context

As opposed to typical ML approaches, GP is very generic

Arbitrary programming language, arbitrary input and output representation
The syntax and semantic of the programming language of consideration serve as
means to provide the algorithm with prior knowledge
(common sense knowledge, background knowledge, domain knowledge).
GP is not the only approach to program induction (but probably the best one :)
See, e.g., inductive logic programming, ILP
GP embodies the ultimate goal of AI: to build a system capable of
self-programming (adaptation, learning).

Summary of our rst glimpse at GP 26

 GP's founding father

http://www.genetic-programming.com/johnkoza.html

Summary of our rst glimpse at GP 27

 Life demonstration of GP using ECJ

Life demonstration of GP using ECJ 28

 GP on ECJ

ECJ, Evolutionary Computation in Java,

http://cs.gmu.edu/~eclab/projects/ecj/
by Sean Luke, Liviu Panait, Gabriel Balan, Sean Paus, Zbigniew Skolicki, Elena
Popovici, Keith Sullivan, et al.
Probably the most popular freely available framework for EC, with a strong
support for GP
Licensed under Academic Free License, version 3.0
As of December 2013: version 21.
Many other libraries integrate with ECJ.

Life demonstration of GP using ECJ 29

 Selected ECJ features

GUI with charting

Platform-independent checkpointing and logging
Hierarchical parameter les
Multithreading
Mersenne Twister Random Number Generators (compare to:
http://www.alife.co.uk/nonrandom/)
Abstractions for implementing a variety of EC forms.
Prepared to work in a distributed environment (including so-called island model)

Life demonstration of GP using ECJ 30

 GP-related ECJ features

GP Tree Representations
Set-based Strongly-Typed Genetic Programming
Ephemeral Random Constants
Automatically-Dened Functions and Automatically Dened Macros
Multiple tree forests
Six tree-creation algorithms
Extensive set of GP breeding operators
Grammatical Encoding
Eight pre-done GP application problem domains (ant, regression, multiplexer,
lawnmower, parity, two-box, edge, serengeti)

Life demonstration of GP using ECJ 31

 Exemplary run of ECJ

Standard output:
java ec.Evolve -file ./ec/app/regression/quinticerc.params
...
Threads: breed/1 eval/1
Seed: 1427743400
Job: 0
Setting up
Processing GP Types
Processing GP Node Constraints
Processing GP Function Sets
Processing GP Tree Constraints
{-0.13063322286594392,0.016487577414659428},
{0.6533404396941143,0.1402200189629743},
{-0.03750634856569701,0.0014027712093654706},
...
{0.6602806044824949,0.13869498395598084},
Initializing Generation 0
Subpop 0 best fitness of generation: Fitness: Standardized=1.1303205 Adjusted=0.46941292 H
Generation 1
Subpop 0 best fitness of generation: Fitness: Standardized=0.6804932 Adjusted=0.59506345 H
...

Life demonstration of GP using ECJ 32

 Exemplary run: The result

The log le produced by the run:

Generation: 0
Best Individual:
Subpopulation 0:
Evaluated: true
Fitness: Standardized=1.1303205 Adjusted=0.46941292 Hits=10
Tree 0:
(* (sin (* x x)) (cos (+ x x)))
Generation: 1
Best Individual:
Subpopulation 0:
Evaluated: true
Fitness: Standardized=0.6804932 Adjusted=0.59506345 Hits=7
Tree 0:
(* (rlog (+ (- x x) (cos x))) (rlog (- (cos (cos (* x x))) (- x x))))
....

Life demonstration of GP using ECJ 33

 Exemplary run

The log le produced by the run:

Best Individual of Run:
Subpopulation 0:
Evaluated: true
Fitness: Standardized=0.08413165 Adjusted=0.92239726 Hits=17
Tree 0:
(* (* (* (- (* (* (* (* x (sin x)) (rlog
x)) (+ (+ (sin x) x) (- x x))) (exp (* x
(% (* (- (* (* (* (* x x) (rlog x)) (+ (+
(sin x) x) (- x x))) (exp (* x (sin x))))
(sin x)) (rlog x)) (exp (rlog x)))))) (sin
x)) (rlog x)) x) (cos (cos (* (* (- (* (*
(exp (rlog x)) (+ x (* (* (exp (rlog x))
(rlog x)) x))) (exp (* (* (* (- (exp (rlog
x)) x) (rlog x)) x) (sin (* x x))))) (sin
x)) (* x (% (* (- (* (* (* (* x x) (rlog
x)) (+ (+ x (+ (+ (sin x) x) (- x x))) (-
x x))) (exp (* x (sin x)))) (sin x)) (rlog
x)) (exp (rlog x))))) x))))

Life demonstration of GP using ECJ 34

 GP Benchmarks

Problem Denition (formula)

Sextic x 6 − 2x 4 + x 2
Septic x 7 − 2x 6 + x 5 − x 4 + x 3 − 2x 2 + x
Nonic x9 + x8 + x7 + x6 + x5 + x4 + x3 + x2 + x
R1 (x + 1)3 /(x 2 − x + 1)
R2 (x 5 − 3x 3 + 1)/(x 2 + 1)
R3 (x 6 + x 5 )/(x 4 + x 3 + x 2 + x + 1)

Life demonstration of GP using ECJ 35

 Even more GP benchmarks ...

Symbolic Regression Drug Bioavailability [46]

Tower [55] ... Protein Structure Classication [61]
Boolean Functions Time Series Forecasting [56]
N-Multiplexer [18], N-Majority [18], N-Parity [18] Path-nding and Planning
Generalised Boolean Circuits [12, 62] Physical Travelling Salesman [31]
Digital Adder [58] Articial Ant [18]
Order [9] Lawnmower [19]
Digital Multiplier [58] Tartarus Problem [6]
Majority [9] Maximum Overhang [40]
Classication Circuit Design [32]
mRNA Motif Classication [27] Control Systems
DNA Motif Discovery [28] Chaotic Dynamic Systems Control [29]
Intrusion Detection [11] Pole Balancing [35]
Protein Classication [22] Truck Control [17]
Intertwined Spirals [18]
Predictive Modelling
Mackey-Glass Chaotic Time Series [24]
Financial Trading [5, 4, 8]
Sunspot Prediction [18]
GeneChip Probe Performance [25]
Prime Number Prediction [57]

Life demonstration of GP using ECJ 36

 ... and more ....

Game-Playing
TORCS Car Racing [53]
Ms PacMan [10]
Othello [30]
Chessboard Evaluation [47]
Backgammon [47]
Mario [52]
NP-Complete Puzzles [15]
Robocode [47]
Rush Hour [47]
Checkers [47]
Freecell [47]
Dynamic Optimisation
Dynamic Symbolic Regression [38, 39, 54]
Dynamic Scheduling [13]
Traditional Programming
Sorting [16, 1]

Life demonstration of GP using ECJ 37

 A more detailed view on GP
(vanilla GP is not all the story)

A more detailed view on GP, (vanilla GP is not all the story) 38

 Where to get [random] candidate solutions from?

Every stochastic search method relies on some sampling algorithm(s)

The distribution of randomly generated solutions is important, as it implies
certain bias of the algorithm.
Problems:
We don't know the `ideal' distribution of GP programs.
Even if we knew it, it may be dicult to design an algorithm that obeys it.
The most widely used contemporary initialization methods take care only of the
syntax of generated programs.
Mainly: height constraint.

A more detailed view on GP, (vanilla GP is not all the story) 39

 Initialization: Full method

Specify the maximum tree height hmax .

The full method for initializing trees:
Choose nonterminal nodes at random until hmax is reached
Then choose only from terminals.

A more detailed view on GP, (vanilla GP is not all the story) 40

 Initialization: Grow method

Specify the maximum tree height hmax .

The grow method for initializing trees:
Choose nonterminal or terminal nodes at random until hmax is reached
Then choose only from terminals.

A more detailed view on GP, (vanilla GP is not all the story) 41

 Initialization: Ramped half and half method

Ramped half and half

Initialize half of population using Full method
Initialize half of population using Grow method

A more detailed view on GP, (vanilla GP is not all the story) 42

 Homologous crossover for GP

Earliest example: one-point crossover [26]: identify a common region in the

parents and swap the corresponding trees.
The common region is the `intersection' of parent trees.

A more detailed view on GP, (vanilla GP is not all the story) 43

 Uniform crossover for GP

Works similarly to uniform crossover in GAs

The ospring is build by iterating over nodes in the common region and ipping a
coin to decide from which parent should an instruction be copied [41]

A more detailed view on GP, (vanilla GP is not all the story) 44

 How to employ multiple operators for `breeding' ?

How should the particular operators coexist in an evolutionary process? In other words:

How should they be superimposed?

What should be the `piping' of particular breeding pipelines?
A topic surprisingly underexplored in GP (and in EC probably too).
An example: Which is better:
pop.subpop.0.species.pipe = ec.gp.koza.MutationPipeline
pop.subpop.0.species.pipe.num-sources = 1
pop.subpop.0.species.pipe.source.0 = ec.gp.koza.CrossoverPipeline

Or:
pop.subpop.0.species.pipe.num-sources = 2
pop.subpop.0.species.pipe.source.0 = ec.gp.koza.CrossoverPipeline
pop.subpop.0.species.pipe.source.0.prob = 0.9
pop.subpop.0.species.pipe.source.1 = ec.gp.koza.MutationPipeline
pop.subpop.0.species.pipe.source.1.prob = 0.1

A more detailed view on GP, (vanilla GP is not all the story) 45

 The Challenges for GP

The Challenges for GP 46

 Bloat

The evolving expressions tend to grow indenitely in size.

For tree-based representations, this growth is typically exponential[-ish]
Evaluation becomes slow, algorithm stalls, memory overrun likely.
One of the most intensely studied topics in GP: 240+ papers as of March, 2012.

The Challenges for GP 47

 Bloat example

Average number of nodes per generation in a typical run of GP solving the Sextic
problem x 6 − 2x 4 + x 2 .

(GP: dotted line)

The Challenges for GP 48

 Countermeasures for bloat

Constraining tree height

Surprisingly, can speed up bloat!
Favoring small programs:
Lexicographic parsimony pressure: given two equally t individuals, prefer (select)
the one represented by a smaller tree.
Bloat-aware operators: size-fair crossover.

The Challenges for GP 49

 Highly non-uniform distribution of program `behaviors'
Convergence of binary Boolean random linear functions (composed of AND, NAND,
OR, NOR, 8 bits)

From: [23] Langdon, W. B. Cantú-Paz, E. (ed.) Random Search is Parsimonious Late

Breaking Papers at the Genetic and Evolutionary Computation Conference
(GECCO-2002), AAAI, 2002, 308-315
The Challenges for GP 50
 High cost of evaluation

Running a program on multiple inputs

can be expensive.
Particularly for some types of data,
e.g., images
Solutions:
Caching of outcomes of subprograms
Parallel execution of programs on
particular tness cases
Bloat prevention methods

The Challenges for GP 51

 Variants of GP

Variants of GP 52
Strongly typed GP (STGP)

A way to incorporate prior knowledge and impose a structure on programs [34]

Implementation:
Provide a set of types
For each instruction, dene the types of its arguments and outcomes
Make the operators type-aware:
Mutation: substitute a random tree of a proper type
Crossover: swap trees of compatible1 types

1 Compatible: belonging to the same `set type'

 Strongly typed GP in ECJ

For the problem of simple classiers represented as decision trees:

Classier syntax: gp.nc.size = 4
Classier ::= Class_id gp.nc.0 = ec.gp.GPNodeConstraints
Classier ::= if_then_else(Condition, Classier, gp.nc.0.name = ncSimpleClassier
Classier) gp.nc.0.returns = class
Condition ::= Input_Variable = Constant_Value gp.nc.0.size = 0
gp.nc.1 = ec.gp.GPNodeConstraints
Implementation in ECJ parameter les: gp.nc.1.name = ncCompoundClassier
gp.type.a.size = 3 gp.nc.1.returns = class
gp.type.a.0.name = class gp.nc.1.size = 4
gp.type.a.1.name = var gp.nc.1.child.0 = var
gp.type.a.2.name = const gp.nc.1.child.1 = const
gp.type.s.size = 0 gp.nc.1.child.2 = class
gp.tc.size = 1 gp.nc.1.child.3 = class
gp.tc.0 = ec.gp.GPTreeConstraints gp.nc.2 = ec.gp.GPNodeConstraints
gp.tc.0.name = tc0 gp.nc.2.name = ncVariable
gp.tc.0.fset = f0 gp.nc.2.returns = var
gp.tc.0.returns = class gp.nc.2.size = 0
gp.nc.3 = ec.gp.GPNodeConstraints
gp.nc.3.name = ncConstant
gp.nc.3.returns = const
gp.nc.3.size = 0
 Linear Genetic Programming

Motivation:
Tree-like structures are not natural for contemporary hardware architectures
Program = a sequence of instructions
Data passed via registers
Pros:
Directly portable to machine code, fast execution.
Natural correspondence to standard (GA-like) crossover operator.
Applications: direct evolution of machine code [36].

Initial register Final register

contents contents

r1 x1 g1 r1

r2 x2 O1 O2 O3 O4 g2 r2

r3 x3 g3 r3

Variants of GP 55
 Linear GP

Variants of GP 56
 Stack-based GP

The best-known representative: Push and PushGP

hampshire.edu/lspector/push.html [51]
Pros:
Very simple syntax: program ::= instruction | literal | ( program* )
No need to specify the number of registers
The top element of a stack has the natural interpretation of program outcome
Natural possibility of implementing autoconstructive programs [50]
Includes certain features that make it Turing-complete (e.g., YANK instruction).

Variants of GP 57
 Push: Example

Program:
( 2 3 INTEGER.* 4.1 5.2 FLOAT.+ TRUE FALSE BOOLEAN.OR )
Initial stack states:
BOOLEAN STACK: ()
CODE STACK: ( 2 3 INTEGER.* 4.1 5.2 FLOAT.+ TRUE FALSE BOOLEAN.OR )
FLOAT STACK: ()
INTEGER STACK: ()
Stack states after program execution:
BOOLEAN STACK: ( TRUE )
CODE STACK: ( ( 2 3 INTEGER.* 4.1 5.2 FLOAT.+ TRUE FALSE BOOLEAN.OR ) )
FLOAT STACK: ( 9.3 )
INTEGER STACK: ( 6 )
http://hampshire.edu/lspector/push3-description.html

Variants of GP 58
 Other variants of GP

Grammatical Evolution: The grammar of the programming language of

consideration is given as input to the algorithm. Individuals encode the choice of
productions in the derivation tree (which of available alternative production
should be chosen, modulo the number of productions available at given step of
derivation).

Graph-based GP
Motivation: standard GP cannot reuse subprograms (within a single program)
Example: Cartesian Genetic Programming

Variants of GP 59
 Other variants of GP

Multiobjective GP. The extra objectives can:

Come with the problem
Result from GP's specics: e.g., use program size as the second (minimized)
objective
Be associated with dierent tests (e.g., feature tests [43])
Developmental GP (e.g., using Push)
Probabilistic GP (a variant of EDA, Estimation of Distribution Algorithms):
The algorithm maintains a probability distribution P instead of a population
Individuals are generated from P `on demand'
The results of individuals' evaluation are used to update P

Variants of GP 60
 Simple EDA-like GP: PIPE
Probabilistic Incremental Program Evolution [44]

Variants of GP 61
 Applications of GP

Applications of GP 62
 Review

GP produced a number of solutions that are human-competitive, i.e., a GP

algorithm automatically solved a problem for which a patent exists2 .
A recent award-winning work has demonstrated the ability of a GP system to
automatically nd and correct bugs in commercially-released software when
provided with test data3 .
GP is one of leading methodologies that can be used to `automate' science,
helping the researchers to nd the hidden complex patterns in the observed
phenomena4 .

2 Koza, J. R., Keane, M. A., Streeter, M. J., Mydlowec, W., Yu, J., Lanza, G., 2003. Genetic Pro-
gramming IV: Routine Human-Competitive Machine Intelligence. Kluwer Academic Publishers.
3 Arcuri, A., Yao, X., A novel co-evolutionary approach to automatic software bug xing. In: Wang,
J. (Ed.), 2008 IEEE World Congress on Computational Intelligence. IEEE Computational Intelligence
Society, IEEE Press, Hong Kong.
4 Schmidt, M., Lipson, H., 3 Apr. 2009. Distilling free-form natural laws from experimental data.
Science 324 (5923), 8185.
Applications of GP 63
 Humies

(...) Entries were solicited for cash awards for human-competitive results that were
produced by any form of genetic and evolutionary computation and that were published

Applications of GP 64
 Humies

The conditions to qualify:

(A) 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.
(B) The result is equal to or better than a result that was accepted as a new scientic
result at the time when it was published in a peer-reviewed scientic journal.
(C) The result is equal to or better than a result that was placed into a database or
archive of results maintained by an internationally recognized panel of scientic
experts.
(D) The result is publishable in its own right as a new scientic result  independent
of the fact that the result was mechanically created.
(E) 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.
(F) The result is equal to or better than a result that was considered an achievement
in its eld at the time it was rst discovered.
(G) The result solves a problem of indisputable diculty in its eld.
(H) The result holds its own or wins a regulated competition involving human
contestants (in the form of either live human players or human-written computer
programs).

Applications of GP 65
 Selected Gold Humies

2004: Jason D. Lohn Gregory S. Hornby Derek S. Linden, NASA Ames Research
Center,
An Evolved Antenna for Deployment on NASA's Space Technology 5 Mission

http://idesign.ucsc.edu/papers/hornby_ec11.pdf

Applications of GP 66
 Selected Gold Humies using GP

2009: Stephanie Forrest Claire Le Goues ThanhVu Nguyen Westley Weimer

Automatically nding patches using genetic programming: A Genetic
Programming Approach to Automated Software Repair

Applications of GP 67
 Selected Gold Humies using GP

2008: Lee Spector David M. Clark Ian Lindsay Bradford Barr Jon Klein
Genetic Programming for Finite Algebras
2010: Natalio Krasnogor Paweª Widera Jonathan Garibaldi
Evolutionary design of the energy function for protein structure prediction GP
challenge: evolving the energy function for protein structure prediction Automated
design of energy functions for protein structure prediction by means of genetic
programming and improved structure similarity assessment
2011: Achiya Elyasaf Ami Hauptmann Moshe Sipper
GA-FreeCell: Evolving Solvers for the Game of FreeCell

Applications of GP 68
 Other applications

classication problems in machine learning [20], object recognition [21, 37], or

learning game strategies [14] .
[2, 60] has demonstrated the ability of a GP system to automatically nd and
correct bugs in commercially-released software when provided with test data.
In the context of this paper, it deserves particular attention that GP is one of
leading methodologies that can be used to `automate' science, helping the
researchers to nd the hidden complex patterns in the observed phenomena.
In this spirit, in their seminal paper [45] have shown how GP can be used to
induce scientic laws from experimental data. Many other studies have
demonstrated the usefulness of GP for modeling dierent phenomena, including
those of natural origins [48, 3, 7, 59, 49].

See [42] for an extensive review of GP applications.

Applications of GP 69
 Additional resources

Additional resources 70
 Additional resources

Evolutionary Computation in Java cs.gmu.edu/~eclab/projects/ecj/

Generic software framework for EA, well-prepared to work with GP
The online GP bilbiography www.cs.bham.ac.uk/~wbl/biblio/

The genetic programming `home page' (a little bit messy, but still valuable)
http://www.genetic-programming.com/

Additional resources 71
 Bibliography

Bibliography 72
 A. Agapitos and S. M. Lucas.
Evolving Modular Recursive Sorting Algorithms.
In Proc. EuroGP, 2007.
A. Arcuri and X. Yao.
A novel co-evolutionary approach to automatic software bug xing.
In J. Wang, editor, 2008 IEEE World Congress on Computational Intelligence,
pages 162168, Hong Kong, 1-6 June 2008. IEEE Computational Intelligence
Society, IEEE Press.
M. Arganis, R. Val, J. Prats, K. Rodriguez, R. Dominguez, and J. Dolz.
Genetic programming and standardization in water temperature modelling.
Advances in Civil Engineering, 2009, 2009.
A. Brabazon, M. O'Neill, and I. Dempsey.
An Introduction to Evolutionary Computation in Finance.
IEEE Computational Intelligence Magazine, 3(4):4255, 2008.
R. Bradley, A. Brabazon, and M. O'Neill.
Dynamic High Frequency Trading: A Neuro-Evolutionary Approach.
In Proc. EvoWorkshops, 2009.
G. Cuccu and F. Gomez.
When novelty is not enough.
In Proc. EvoApplications, 2011.
Bibliography 73
 M. Daga and M. C. Deo.
Alternative data-driven methods to estimate wind from waves by inverse
modeling.
Natural Hazards, 49(2):293310, May 2009.
I. Dempsey, M. O'Neill, and A. Brabazon.
Adaptive Trading With Grammatical Evolution.
In Proc. CEC, 2006.
G. Durrett, F. Neumann, and U.-M. O'Reilly.
Computational Complexity Analysis of Simple Genetic Programming On Two
Problems Modeling Isolated Program Semantics.
In Proc. FOGA, 2011.
E. Galván-López, J. Swaord, M. O'Neill, and A. Brabazon.
Evolving a Ms. PacMan Controller Using Grammatical Evolution.
In Applications of Evolutionary Computation. Springer, 2010.
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, 43:13621374, 2007.

Bibliography 74
 S. Harding, J. F. Miller, and W. Banzhaf.
Developments in Cartesian Genetic Programming: self-modifying CGP.
GPEM, 11:397439, 2010.
D. Jakobovi¢ and L. Budin.
Dynamic Scheduling with Genetic Programming.
In Proc. EuroGP, 2006.
W. Ja±kowski, K. Krawiec, and B. Wieloch.
Evolving strategy for a probabilistic game of imperfect information using genetic
programming.
Genetic Programming and Evolvable Machines, 9(4):281294, 2008.
G. Kendall, A. Parkes, and K. Spoerer.
A Survey of NP-Complete Puzzles.
International Computer Games Association Journal, 31(1):1334, 2008.
K. E. Kinnear, Jr.
Evolving a Sort: Lessons in Genetic Programming.
In Proc. of the International Conference on Neural Networks, 1993.
J. Koza.
A Genetic Approach to the Truck Backer Upper Problem and the Inter-twined
Spiral Problem.
In Proc. International Joint Conference on Neural Networks, 1992.
Bibliography 75
 J. R. Koza.
Genetic Programming: On the Programming of Computers by Means of Natural
Selection.
MIT Press, Cambridge, MA, USA, 1992.
J. R. Koza.
Genetic Programming II: Automatic Discovery of Reusable Programs.
MIT Press, Cambridge Massachusetts, May 1994.
K. Krawiec.
Genetic programming-based construction of features for machine learning and
knowledge discovery tasks.
Genetic Programming and Evolvable Machines, 4:329343, 2002.
K. Krawiec and B. Bhanu.
Visual learning by evolutionary and coevolutionary feature synthesis.
IEEE Trans. on Evolutionary Computation, 11:635650, October 2007.
DOI: 10.1109/TEVC.2006.887351.
W. Langdon and W. Banzhaf.
Repeated Patterns in Genetic Programming.
Natural Computing, 7:589613, 2008.

Bibliography 76
 W. B. Langdon.
Random search is parsimonious.
In E. Cantú-Paz, editor, Late Breaking Papers at the Genetic and Evolutionary
Computation Conference (GECCO-2002), pages 308315, New York, NY, 9-13
July 2002. AAAI.
W. B. Langdon and W. Banzhaf.
Repeated Sequences in Linear Genetic Programming Genomes.
Complex Systems, 15(4):285306, 2005.
W. B. Langdon and A. P. Harrison.
Evolving Regular Expressions for GeneChip Probe Performance Prediction.
In Proc. PPSN, pages 10611070, 2008.
W. B. Langdon and R. Poli.
Foundations of Genetic Programming.
Springer-Verlag, 2002.
W. B. Langdon, J. Rowsell, and A. P. Harrison.
Creating Regular Expressions as mRNA Motifs with GP to Predict Human Exon
Splitting.
In Proc. GECCO, 2009.

Bibliography 77
 W. B. Langdon, O. Sanchez Graillet, and A. P. Harrison.
Automated DNA Motif Discovery.
arXiv.org, 2010.
M. Lones, A. Tyrrell, S. Stepney, and L. Caves.
Controlling Complex Dynamics with Articial Biochemical Networks.
In Proc. EuroGP, pages 159170, 2010.
S. Lucas.
Othello Competition.
http:/\protect\kern-.1667em\relax/algoval.essex.ac.uk:
8080/othello/html/Othello.html, 2012.
[Online; accessed 27-Jan-2012].
S. Lucas.
The Physical Travelling Salesperson Problem.
http:
/\protect\kern-.1667em\relax/algoval.essex.ac.uk/ptsp/ptsp.html,
2012.
[Online: accessed 27Jan-2012].
T. McConaghy.
FFX: Fast, Scalable, Deterministic Symbolic Regression Technology.
In Proc. GPTP, 2011.

Bibliography 78
 T. M. Mitchell.
Machine Learning.
McGraw-Hill, 1997.
D. J. Montana.
Strongly typed genetic programming.
BBN Technical Report #7866, Bolt Beranek and Newman, Inc., 10 Moulton
Street, Cambridge, MA 02138, USA, 7 May 1993.
M. Nicolau, M. Schoenauer, and W. Banzhaf.
Evolving Genes to Balance a Pole.
In Proc. EuroGP, 2010.
P. Nordin and W. Banzhaf.
Genetic programming controlling a miniature robot.
In E. V. Siegel and J. R. Koza, editors, Working Notes for the AAAI Symposium
on Genetic Programming, pages 6167, MIT, Cambridge, MA, USA, 1012 Nov.
1995. AAAI.
G. Olague and L. Trujillo.
Evolutionary-computer-assisted design of image operators that detect interest
points using genetic programming.
Image and Vision Computing, In Press, Accepted Manuscript, 2011.

Bibliography 79
 M. O'Neill, A. Brabazon, and E. Hemberg.
Subtree Deactivation Control with Grammatical Genetic Programming in
Dynamic Environments.
In Proc. CEC, 2008.
M. O'Neill and C. Ryan.
Grammatical Evolution by Grammatical Evolution: The Evolution of Grammar
and Genetic Code.
In Proc. EuroGP, pages 138149. Springer-Verlag, 57 Apr. 2004.
M. Paterson, Y. Peres, M. Thorup, P. Winkler, and U. Zwick.
Maximum Overhang.
In Proc. 19th Annual ACM-SIAM Symposium on Discrete Algorithms, 2008.
R. Poli and W. B. Langdon.
On the search properties of dierent crossover operators in genetic programming.
In 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, pages 293301, University of
Wisconsin, Madison, Wisconsin, USA, 22-25 July 1998. Morgan Kaufmann.

Bibliography 80
 R. Poli, W. B. Langdon, and N. F. McPhee.
A eld 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).
B. J. Ross and H. Zhu.
Procedural texture evolution using multiobjective optimization.
New Generation Computing, 22(3):271293, 2004.
R. P. Salustowicz and J. Schmidhuber.
Probabilistic incremental program evolution.
Evolutionary Computation, 5(2):123141, 1997.
M. Schmidt and H. Lipson.
Distilling free-form natural laws from experimental data.
Science, 324(5923):8185, 3 Apr. 2009.
S. Silva and L. Vanneschi.
State-of-the-Art Genetic Programming for Predicting Human Oral Bioavailability
of Drugs.
In Proc. 4th International Workshop on Practical Applications of Computational
Biology and Bioinformatics, 2010.

Bibliography 81
 M. Sipper.
Let the Games Evolve!
In Proc. GPTP, 2011.
C. Sivapragasam, R. Maheswaran, and V. Venkatesh.
Genetic programming approach for ood routing in natural channels.
Hydrological Processes, 22(5):623628, 2007.
C. Sivapragasam, N. Muttil, S. Muthukumar, and V. M. Arun.
Prediction of algal blooms using genetic programming.
Marine Pollution Bulletin, 60(10):18491855, 2010.
L. Spector.
Towards practical autoconstructive evolution: Self-evolution of problem-solving
genetic programming systems.
In R. Riolo, T. McConaghy, and E. Vladislavleva, editors, Genetic Programming
Theory and Practice VIII, volume 8 of Genetic and Evolutionary Computation,
chapter 2, pages 1733. Springer, Ann Arbor, USA, 20-22 May 2010.
L. Spector, C. Perry, and J. Klein.
Push 2.0 programming language description.
Technical report, School of Cognitive Science, Hampshire College, Apr. 2004.

Bibliography 82
 J. Togelius, S. Karakovskiy, J. Koutnik, and J. Schmidhuber.
Super Mario Evolution.
In Proc. IEEE Computational Intelligence and Games, 2009.
TORCS: The Open Car Racing Simulator.
http://torcs.sourceforge.net/, 2012.
L. Vanneschi and G. Cuccu.
Variable Size Population for Dynamic Optimization with Genetic Programming.
In Proc. GECCO, 2009.
E. Vladislavleva, G. Smits, and D. Den Hertog.
Order of Nonlinearity as a Complexity Measure for Models Generated by Symbolic
Regression via Pareto Genetic Programming.
IEEE Trans EC, 13(2):333349, 2009.
N. Wagner, Z. Michalewicz, M. Khouja, and R. McGregor.
Time Series Forecasting for Dynamic Environments: The DyFor Genetic Program
Model.
IEEE Trans EC, 2007.
J. Walker and J. Miller.
Predicting Prime Numbers Using Cartesian Genetic Programming.
In Proc. EuroGP, 2007.

Bibliography 83
 J. A. Walker, K. Völk, S. L. Smith, and J. F. Miller.
Parallel Evolution using Multi-chromosome Cartesian Genetic Programming.
GPEM, 10:417445, 2009.
W.-C. Wang, K.-W. Chau, C.-T. Cheng, and L. Qiu.
A comparison of performance of several articial intelligence methods for
forecasting monthly discharge time series.
Journal of Hydrology, 374(3-4):294306, 2009.
W. Weimer, S. Forrest, C. Le Goues, and T. Nguyen.
Automatic program repair with evolutionary computation.
Communications of the ACM, 53(5):109116, June 2010.
P. Widera, J. Garibaldi, and N. Krasnogor.
GP challenge: Evolving energy function for protein structure prediction.
GPEM, 11:6188, 2010.
T. Yu.
Hierarchical Processing for Evolving Recursive and Modular Programs Using
Higher-Order Functions and Lambda Abstraction.
GPEM, 2:345380, 2001.

Bibliography 84