https://www.studyc.

info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/262185321

Physiological Systems Modeling, Simulation, and

Control

Chapter · June 2012

DOI: 10.4018/978-1-4666-0122-2.ch017

CITATIONS READS

0 1,708

2 authors:

Konstantinos Michmizos Konstantina Nikita

Rutgers, The State University of New Jersey National Technical University of Athens
66 PUBLICATIONS 227 CITATIONS 429 PUBLICATIONS 4,388 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Deep Brain Stimulation (DBS) View project

master research View project

All content following this page was uploaded by Konstantinos Michmizos on 11 May 2014.

The user has requested enhancement of the downloaded file.

https://www.studyc.info/
Handbook of Research on
Biomedical Engineering
Education and Advanced
Bioengineering Learning:
Interdisciplinary Concepts
Ziad O. Abu-Faraj
American University of Science and Technology, Lebanon

Volume I

https://www.studyc.info/
Managing Director: Lindsay Johnston
Senior Editorial Director: Heather Probst
Book Production Manager: Sean Woznicki
Development Manager: Joel Gamon
Acquisitions Editor: Erika Gallagher
Typesetter: Jennifer Romanchak
Cover Design: Nick Newcomer, Lisandro Gonzalez

Published in the United States of America by

Medical Information Science Reference (an imprint of IGI Global)
701 E. Chocolate Avenue
Hershey PA 17033
Tel: 717-533-8845
Fax: 717-533-8661
E-mail: [email protected]
Web site: http://www.igi-global.com

Copyright © 2012 by IGI Global. All rights reserved. No part of this publication may be reproduced, stored or distributed in
any form or by any means, electronic or mechanical, including photocopying, without written permission from the publisher.
Product or company names used in this set are for identification purposes only. Inclusion of the names of the products or
companies does not indicate a claim of ownership by IGI Global of the trademark or registered trademark.

Library of Congress Cataloging-in-Publication Data

Biomedical engineering education and advanced bioengineering learning: interdisciplinary concepts / Ziad O. Abu-Faraj,
editor.
p. cm.
Includes bibliographical references and index.
Summary: “This book explores how healthcare practices have been steered toward emerging frontiers, including, among
others, functional medical imaging, regenerative medicine, nanobiomedicine, enzyme engineering, and artificial sensory
substitution”-- Provided by publisher.
ISBN 978-1-4666-0122-2 (hardcover) -- ISBN 978-1-4666-0123-9 (ebook) -- ISBN 978-1-4666-0124-6 (print & perpetual
access) 1. Biomedical engineering--Study and teaching. I. Abu-Faraj, Ziad O., 1964-
R856.3.B56 2012
610.28076--dc23
2011042007

British Cataloguing in Publication Data

A Cataloguing in Publication record for this book is available from the British Library.

All work contributed to this book is new, previously-unpublished material. The views expressed in this book are those of the
authors, but not necessarily of the publisher.

https://www.studyc.info/
 745

Chapter 17
Physiological Systems Modeling,
Simulation, and Control
Konstantina S. Nikita
National Technical University of Athens, Greece

Konstantinos P. Michmizos
Massachusetts Institute of Technology, USA

ABSTRACT
Physiological systems modeling, simulation, and control is a research area integrating science and
engineering and contributes to a continuous refinement of knowledge on how the body works. The roots
of modeling a body area date back thousands of years, yet it was not until the 1950s that the tree of
knowledge started to be fed with data-driven hypotheses and interventions. This chapter tries to orga-
nize disparate information of the most important modeling, simulation, and control perspectives into
a coherent set of views currently applied to modern biological and medical research. It is addressed
to researchers on human system physiological modeling, working both in academia and in industry to
address current and future research goals.

17.1. CHAPTER OBJECTIVES time-varying or time-constant variables. Although

it is not possible to cover all of the physiological
It is Zeus anathema on physiological models to modeling domains in the subsequent pages, we
agonize between the Scylla of simulating a biologi- have made an effort to present and briefly discuss
cal system and the Charybdis of controlling such the major fields of activity in which models of
systems. This chapter aims to serve as an introduc- biological systems are engaged. We first provide
tion to and overview of the interdisciplinary field an introduction to important concepts and then we
of modeling, simulation, and control of physi- illustrate these ideas with examples acquired from
ological systems. Research and applications in the physiological systems. We focus on techniques
area extend from cells to organs and systems, and in modeling that motivate the inclusion of con-
include linear and nonlinear approaches having trol mechanisms into physiological systems and

DOI: 10.4018/978-1-4666-0122-2.ch017

Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

models. In parallel, we provide methodological can make predictions before any intervention or
approaches and we discuss their advantages and after failures (lesions). Models can also be used to
limitations in order to motivate the reader to have evaluate the functional limits of an operation, be
a hands-on experience on the main modeling it biological or that of an instrument interrelated
aspects covered. with a biological system. They can also explore
linear behavior at selected operating points. Lastly,
physiological models provide the means (simula-
17.2. INTRODUCTION tions) to truly explore the non-linear nature of the
biological physics.
How does an organ work? What is really happen-
ing inside a diseased organ? How can we monitor
and supervise a drug molecule to help an organ 17.3. COMPREHENSIVE
work in a healthy manner? What is a healthy DEFINITION OF PHYSIOLOGICAL
manner of living for a cell, organ or body any- SYSTEMS MODELING,
ways? The motivation of modeling is convoluted SIMULATION, AND CONTROL
with our distinctive characteristic of wondering.
Models in physiology are mainly used for insight, In order to start thinking about modeling a system,
description, and control. We want to know, and let us begin with the parable of a Saturday theater
sometimes we need to learn, how the components that is crammed to suffocation by all kinds of
of a system and their interconnections generate spectators. By the end of the theatrical play, each
the overall operating characteristics of that sys- of the spectators is asked to talk about his/her ex-
tem. We also seek to capture the characteristics perience. One person, sitting in the last rows of the
of a physiological system response accurately theater finds that the stage design was ingenious.
and concisely. Another, having the opportunity to sit in the first
In practice, the physiological modeling road row of the theater is amazed by the expressiveness
does not resemble the directional straightness of of the actors. A third, positioned in a corner of the
a roman road. Biological signals are typically theater shows a tendency to talk only for specific
amplitude limited and distorted by colored (i.e., scenes of the play; the ones performed near his/
non-white) noise. Signal recordings have limited her side. Each person, inside the theater, gives a
length and are generally nonstationary; whereas, different description of the same object; yet none
the underlying system is either unknown or very keeps the ultimate truth in his/her hands.
complex to describe. But we still need models, Before projecting our parable to modeling,
since they can verify our designs before the pro- three aspects need to be further discussed. First,
totype stage; and, even if they are not exactly ac- in the real world, we cannot go into the mind of
curate, they can help us gain a basic understanding the director. We do not know the script or even
of the underlying system. Models of physiological the number of the actors in the play; and what is
systems often aid in the specification of design more, there is no unbiased observer that holds an
criteria for the design of procedures aimed at unconditional truth. Second, all spectators formed
alleviating pathological conditions. Models also a personal opinion based on a hypothesis of the
summarize the physiological behavior of a system play that was consistent with the data they col-
concisely, making them an appropriate testing bed lected. This activity, which seems easy and natural
for a plethora of scientific hypotheses being stated. to humans, is called abduction. Third, abduction
This has also been proven useful in the design is not an infallible way for discovering truth.
of medical devices. In a clinical setting, models This chapter describes most of the basic tools

746
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

that can be used to create a quantitative formula (e.g., blood or biopsy). Most medically important
of the description abducted from observations on measurands can be grouped into five categories:
physiological systems. i) biopotential (e.g., electromyography - EMG,
In essence, each spectator created a model (a electrocardiography - ECG), ii) pressure flow
descriptive version) of a system (the play). Let us displacement (e.g., velocity, acceleration, force),
introduce some main terminology at this point. A iii) impedance, iv) temperature, and v) chemical
system may be considered to be any collection of concentration.
interconnected processes and/or objects. A model Various factors complicate the choice of bio-
is a representation that approximates the behavior logical input and output (I/O) measurands. First,
of an actual system. This representation is descrip- most of the parameters that are measured in prac-
tive in a certain level of detail, for that system. By tice are quite small as compared with non-medical
using a set of simplifying assumptions, the system parameters in most industries. For example, most
is conceptually reduced to that of a mathematical voltages are in the micro-volt range, and the sig-
model. Therefore, the results of each model have nals are in the audio-frequency range or below.
significant limitations and are valid only in the Many crucial I/O variables in living systems are
regimes of the real world where the assumptions also inaccessible because the proper measurand
are valid. A model is always connected to an ex- transducer interface cannot be achieved without
periment from which we obtain data. To optimize compromising the system (e.g., cardiac output).
the experiment, we need to have access to the data Patient’s comfort is another parameter selection
related to important variables of the model. Con- factor that is also related to the level of invasiveness
sequently, designing and executing an experiment and the safety of the patient in general. Compat-
is a crucial step in modeling that usually involves ibility with existing equipment and the cost of the
a careful and usually time-consuming selection of experiment also affect decisions on the level of
the model’s variables. Next, we will discuss the abduction used to define a physiological model.
two most important classes of variables for any Thus, there are times that a model is forced to be
modeled system: the input and the output. designed with less details as compared to what
The input of a system is the information or was the initial target.
signals that flow into a system, and which can Desired inputs are the physiological signals
normally be manipulated independently. The (i.e., the measurands) that the model is designed
output of the system is the information or sig- to process. In practice, they are subjected to
nals that flow out of a system, and result from two unwanted artifacts; namely, interfering and
the operation of the system on the input. Both modifying inputs. Interfering inputs relate to how
the input and the output can be material flows, things are measured. They are quantities that
voltages, temperatures, pressures, or any other inadvertently affect the data as a consequence
biological signal. The information that is getting of the principles used to acquire and process the
into the system or out of it is depicted by a physi- desired inputs. Modifying inputs relate to how
cal quantity, property or condition that is being the experiment is physically built or laid out.
measured (i.e., the biological signal), usually They are undesired quantities that indirectly af-
called a measurand. In terms of a physiological fect the input by altering the performance of the
system, there are various measurand accessibility measurement itself. They can influence both the
sites, namely a) internal to the body (e.g., blood desired and the interfering inputs. Some undesir-
pressure), b) external to the body (e.g., electrocar- able quantities can act as both a modifying input
diogram potential), c) emanating from the body and an interfering input.
(e.g., infrared radiation), or d) extracted tissue

747
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

A model needs to be tested on some quantitative Physiological modeling is initiated by ex-

measures that describe the goodness of fit between perimental observations of a phenomenon that
the simulated and the true data. The accuracy of lead to a guesstimate or a verbal description of
a single model is the difference between the true the observed system. An initial hypothesis is
value and the predicted value. The difference is formed followed by a mathematical or computa-
sometimes divided by the true value of the quantity tional model that describes our understanding of
measured; this ratio is often expressed as a percent. the phenomenon. The accuracy of the model is
However, the true value of the reference is seldom tested by acquiring some more data and testing
available. The precision of a measurement system, (simulating) the model against the new data. If
also known as reproducibility or repeatability, the model performs adequately, the model is ready
expresses the closeness of the system’s output in to serve its purpose (e.g., to replace a module of
several measurement experiments made in the a control system). If the model’s accuracy does
same way. Typically, this value is determined by not meet performance specifications, then we
statistical analysis of repeated measurements. It need to refine the model. Additional experiments
is related to the number of significant figures to are carried out to acquire even more data and use
which a measurement can be made (e.g., an output them to update our model. Usually, some of the
variable of 2.434 V is more precise than 2.43 V). variables in the model are observable and some
High precision does not imply high accuracy be- are not. Hence, the new experiments aim to pro-
cause precision makes no comparison to the true vide the data that are needed in order to increase
value. Figure 1 illustrates the difference between our understanding of the physiological system.
accuracy and precision. The new data include information about previ-
ously unobservable variables. The process of

Figure 1. Accuracy vs. precision of a model’s output

748
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

Figure 2. Model refinement graph

refining the model using new data continues
until a satisfactory model is obtained. Typically,
a quantitative criterion is used to test the goodness
of fit between the model and the data. One of the
characteristics of a good model is how well it
predicts the future performance of the physiolog-
ical system. The process is illustrated in Figure
2.
Instead of a concluding remark, two important
modeling principles are underlined: i) The starting
point for successful physiological modeling is
always a simple model that gains a basic under-
standing of the underlying system. If that model
partially succeeds in capturing the known or
anticipated behavior, then the subsequent job is
to refine it. ii) An otherwise hidden structure of
a biological process can become clearer if the
process is successfully modeled with adequate
mathematical and statistical concepts. A deep
knowledge of the modeled structure, and of the
way its mathematical representation responds to
change, allows the formulation of hypotheses and
the testing of theories that are usually not evident
from the phenomenological descriptions of the
system. Engineers and scientists aiming to that has gone awry due to disease, trauma or some
model very complex behaviors, such as bio- other intervention, and iii) assist if they are used
medical phenomena, should not escape the to substitute a diminished or lost function (e.g.,
memory of these hallmark principles. robotic systems that help the paretic side of a
patient after a stroke, for example see (Krebs &
17.3.1. Diagnostic and Hogan, 2006), or cardiac pacemaker able to predict
Therapeutic Challenges and control rhythmic heart beats). Usually, these
models have life-supporting or life-sustaining
The results of medical or biological models serve applications.
three different purposes: i) to understand; to have
a deep, profound knowledge of a real physiologi-
cal system, ii) to predict; to know the future of 17.4. HISTORICAL BACKGROUND
such a system that is currently unknown, and iii) AND LITERATURE OVERVIEW
to control; to constrain or manipulate a system
to function inside desirable working conditions. 17.4.1. History of Modeling
In analogy to the above purposes, physiological
models can contribute in i) diagnosis if they ac- The process of modeling a physiological system
quire information for presentation to the human has a long history interconnected with the history
senses (i.e., extend the human senses), ii) therapy of medicine. It was first introduced as a vague
if they are used to control a physiological process concept with rather philosophical roots; and, after

749
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

centuries, it acquired its scientific entity and a Minor with Alcmaeon’s percepts and Empedocles’
proper name. Modeling of the living world, the concepts about the equilibrium to develop the
universe, has its origins in the sixth century BC humoral theory for human physiology (Longriff,
among the Ionian Greeks of Asia Minor. At that 1989). According to this theory, human beings are
time, it was mainly occupied with speculation modeled to consist of a soul and a body, which
about the cause of the universe, and was associ- contain four humors: blood, phlegm, black and
ated with the name of Thales of Miletus, whose yellow bile; humors that correspond to the four
chief successors (also sixth-century BC Milesians) organs of the body: the heart, the brain, the liver and
were Anaximander and Anaximenes. The material the spleen. These four humors were believed to be
principle of the universe was modeled as a single in continuous motion through the circulation. The
uncreated and imperishable substance that under- equilibrium and the harmony of the four humors
went various modifications to produce the plethora (eucrasia in Greek terminology) were identified
of phenomena in the universe. Thales thought that with health. Their disequilibrium and disharmony
this substance was water; Anaximander defined (dyscrasia in Greek terminology) produces what
it as something indeterminate, without specific is known as disease (Marketos, 1997).
qualities, and Anaximenes believed it was the air. For six consecutive centuries, the Hippocratic
Around 500 BC, Alcmaeon of Croton, a Greek view of humorism that regarded the disease as a
writer and philosopher, localized the brain as the dynamic process, withstood the pressure of the
center of understanding reality and introduced Atomists’ view of the body as an interconnec-
brain pathways by using the term channels (poroi tion of indivisible particles in which the disease
– πόροι) that connected the brain to the sensory remained a static phenomenon. Around 150 AD,
organs. By using a political metaphor, he was Galenos’ understanding of anatomy and medicine,
also the first to relate health with balance. He principally influenced by theory of humorism,
defined a healthy body as the result of equality reestablished the Hippocratic ideas of the unity
(isonomia – ισονομία) of opposing powers (e.g., of the organism in which the interaction with the
hot vs. cold) which make up the body. Empedocles environment (homeostatis) is crucial for survivor.
(490 BC - 430 BC), a Greek philosopher that His theories dominated and influenced Western
lived in Sicily, was the advocate of the segrega- medical science for nearly two millennia. Gale-
tion of the matter to four basic elements: water, nos’ theory of the physiology of the circulatory
earth, air, and fire. He was the first to consider an system endured until 1628, when William Harvey
interconnection among the various compartments published his treatise entitled De motu cordis, in
of his model of the human body. In addition to which he established a model of blood circula-
the four elements (which he called roots), he used tion with the heart acting as a pump (Furley &
the words love (philotis – φιλότις) to model the Wilkie, 1984). Stephen Hales, nearly a century
attraction of different forms of matter, and strife later, introduced arterial elasticity and postulated
(neikos - νείκος) to account for their separation. its buffering effect on the pulsatile nature of blood
He considered love and strife to be distinct sub- flow (Hales, 1733). He modeled the depulsing
stances in equilibrium, with the four elements in effect with the fire engines of his day, in which a
solution with them. chamber with an air-filled dome, “inverted globe”,
Interrupting centuries of superposition and acted to cushion the bolus from the inlet water
mythology that entwined the understanding of pump so that “a more nearly equal spout” flowed
the real world and the treatment of diseases, Hip- out of the nozzle. His analogy became the basis of
pocrates (ca. 460 BC - ca. 370 BC) combined the first modern cardiovascular models. In 1897,
the sixth century BC philosophical trend of Asia Stewart first measured cardiac output in intact

750
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

animals (Stewart, 1897), more or less affirming organisms function. Multidisciplinary scientific
Harvey’s calculations. Krogh and Erlang, in 1919, research spotlights the characteristics of vari-
presented what is believed to be the first paper ous physiological systems. Complex, nonlinear,
on mathematical modeling in biomedical science nonhomogeneous, discontinuous, anisotropic,
(Krogh, 1919). About ten years later, Wiggers multilayered, multidimensional, etc. systems
used Fourier analysis to describe intraventricular needed the development of analogous models
pressure waveforms (Wiggers, 1928). that described them.
In 1952, Alan Lloyd Hodgkin and Andrew Hux-
ley initiated the sub-cellular and cellular modeling. 17.4.2. Evolution of Computer Power
They presented a set of nonlinear ordinary differ- and Relation with Advancements in
ential equations that approximates the electrical Physiological Systems Modeling
characteristics of excitable cells such as neurons
and cardiac myocytes. Their model explains the In the second half of the 20th century, biological
ionic mechanisms underlying the initiation and models, used to describe and classify the normal
propagation of action potentials in the squid giant and abnormal physiological conditions, pushed
axon (Hodgkin & Huxley, 1952). For their work, the researchers to descend the modeling ladder:
they received the 1963 Nobel Prize in Physiol- from the organismal level down to the sub-cellular
ogy or Medicine, and the Hodgkin-Huxley model and even nuclear (gene) level. But before the use
became the “paradigm” physiological model of of digital computers, mathematical models of
nerve excitation. A few years later, Noble presented biomedical systems were either oversimplified
the first cardiac Purkinje fiber cell model (Noble, or involved a great deal of hand calculation as
1960). These two works set the foundations for the described in the Hodgkin-Huxley investiga-
development of the current, quantitative approach tions published in 1952. Since the 1980’s, the
to computational modeling of biological systems, progressive introduction of the digital computer,
which is thoroughly based on experimental data, programming languages, and simulation software
and aims to make experimentally verifiable pre- to every lab space in laboratories across the world
dictions. Together with the first physiological enormously shrank the time required to acquire
models, the methodology to acquire them began data from simulation experiments. In fact, since
to develop as well. The iterative process in model the 1990’s, digital computer environment became
building was first introduced by Popper, who the working place for any scientist; and, the terms
pointed out that no model should be considered modeling and simulation have almost become
perfect. In fact, he proposed that models must synonymous. In addition, the internet boom at
exhibit “falsifiability” (Popper, 1959). the start of the new millennium was the major
During the last 50 years, fuelled both by ad- contributor to the international partnership among
vancements of digital computers, programming scientists, and allowed for time and resource
languages, and simulation software and by the consuming modeling projects to become feasible
increasing demand in quantitative assessment since simulations could run on multiple process-
of element interrelations in physiological sys- ing sites spread throughout the world. This has
tems, computational modeling of physiological allowed the development of much more realistic
processes and systems witnessed a remarkable or homeomorphic models that include as much
development. Now attention is shifting toward knowledge as possible about the structure and
integrative computational modeling in biomedi- interrelationships of the physiological system
cal research to link the magnificent body of new without any overriding concern about the number
knowledge to an understanding of how intact of calculations.

751
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

The development of information-gathering Though models can continue to be made more

technologies and the introduction of modeling complex, it is important to evaluate the value added
methodologies that incorporate large-scale data with each stage of complexity – the model should
have facilitated a dramatic increase in the degree be made as simple as possible to explain the data,
of quantification applied to modern physiological but not so simple that it becomes meaningless. On
research. In the past few years, computational the other hand, a model that is made too complex
modeling and analysis played a critical role in is also of little use. Such models fail to generalize
decoding complex systems descriptions from large well either due to a lack of computing resources
sets of noisy and sometimes redundant data, and (such as time and processing power), or because
in developing an engineering understanding of they are gradually becoming sluggish in keeping
physiological systems. In November 2010, the pace with the new knowledge that is constantly
search-term “modelling OR modeling” yielded being added to the description of physiological
over 111,000 entries in PubMed, with more than systems. Models, currently developing, consti-
58,000 since the year 2006. Thus, almost half tute a pivotal point in solving the many open
of the papers appeared in the last four years, as questions of human systems’ dynamics, and the
compared to the rest of the papers published in the information processing from singe cells. The
preceding six decades. These developments show present and forthcoming advances in biology and
that the distance between theory (models) and systems modeling are expected not only to further
experiment (simulations) is rapidly diminishing. increase the huge amount of information coming
The start of the 21st century has found research- from physiological studies, but also to represent
ers working behind their computers climbing an opportunity to help improve the well-being or
the simulation ladder, and composing low level quick healing of individuals facing health issues.
information to gradually form a first-principles
physiological knowledge from the low scale of the 17.4.3. Presentation of Current
nucleus of a cell all the way to the level of a com- Projects: The Physiome Project, The
plex organism. Various international cooperation Virtual Physiological Human-VPH
projects on healthcare information systems, based
on grid capabilities and biomedical informatics, Whereas the reductionist approach in the last
among European Union (EU), North and Latin century focused on studies of isolated systems
America, and North Africa countries, aim to cre- aiming for the finest possible molecular and cel-
ate a common health information infrastructure in lular events, integration is becoming the most
Western countries and extend it to other regions. popular scientific term today. The remarkable
In EU, various FP6 initiatives such as SHARE achievement of completion of the first draft of
(http://www.eu-share.org), ImmunoGrid (http:// the human genome sequence demonstrates the
www.immunogrid.org), SeaLife (http://www. power of integration of the interdisciplinary
biotec.tu-dresden.de/sealife), and ACGT (http:// scientific power. Following the contemporary
www.eu-acgt.org) have concluded successfully, trend, currently developed models aim not only
and other FP7 initiatives, such as Sim-e-child to explicitly understand the physiological entity
(http://www.sim-e-child.org) and ActioGrid under study, but also to relate the subsystems’
(http://www.action-grid.eu) have begun. At the interconnections to the systemic behavior. Scien-
planet level, HealthGrid initiative (http://initia- tists trawl for relations in a large area extending
tive.healthgrid.org), supported by the HealthGrid from molecules, genes, proteins, cells, organs, and
Association, was created to promote deployment systems up to whole organisms. Interrelationships
of grid technologies in health. among biological systems span more than one

752
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

descriptive level, at all space and time scales. The biology undertaken by individual laboratories
aggregation of various modeling levels is achieved mainly in Western countries. Financial support is
by identifying appropriate variables that can be provided mainly from national and international
omitted, averaged, or approximated. In that sense, health research agencies.
a newly developed model should be placed with The main scope of those projects is to gather
respect to a modeling hierarchy at all scales so that interdisciplinary modeling work, information pro-
parameters in one model are the output of models cessing methodologies and relevant software tools,
at a finer spatial or temporal scale. data banks, etc., and make them approachable to
The elucidation of such multilevel models re- research groups across the globe. However, most of
lies on acquiring detailed structural and functional the projects have just begun and have not achieved
information. For instance, research on Parkinson’s great depth yet, for many theoretical and techno-
disease is based on data ranging from the properties logical issues have to be addressed. The challenge
of membrane ion channels using patch clamp tech- for the projects is to link these two developments
niques, to neuronal in vivo characteristics available for an individual – to use complementary data
by means of multiple microelectrodes, populations together with computational modeling tailored
of neurons using stereo-electroencephalography to the anatomy, physiology and genetics of that
or electro-corticography, up to extended brain ac- individual, for diagnosis or treatment.
tivities with high density electroencephalography The Physiome Project represents current quan-
and magneto-encephalography. titative attempts in this direction that establish
Nevertheless, the knowledge gathered is ham- top-down paths to meet up with the sub-cellular
pered by the system’s intrinsic complexity and information and, so, introduce models traveling
by the fact that biological mechanisms are still the whole way from genes to health. Its concept
poorly understood. That is why model design, was first presented in a report from the Com-
experimental investigations and observational mission on Bioengineering in Physiology to the
tools have to be wisely chosen to represent con- International Union of Physiological Sciences
sistently the true system. Scientists in medicine, (IUPS) Council at the 32nd World Congress in
biology, physics, chemistry, applied mathematics, Glasgow, in 1993. The name of the project comes
and computer and engineering science are needed from “physio-” (φύσις- life) and “-ome” (as a
to collaborate. Database management, recognition whole), and is intended to provide a “quantita-
and fusion of multidimensional signals and sens- tive description of physiological dynamics and
ing devices are to provide the means to modeling functional behaviour of the intact organism”. A
and control studies. synthesium on the Physiome Project was held at
Over the past decade, several model integration the 34th World Congress of IUPS in Christchurch,
initiatives have been launched that aim to create New Zealand, in August 2001, and the Physiome
reliable biological and physiological models, Project was designated as a major focus for IUPS
including projects like E-cell, Virtual Cell, the for the subsequent decades. The main projects of
Virtual Physiological Human, and the Physiome the Physiome include models of the brain and the
Projects. These projects attempt to formulate central nervous system, the cardiovascular, the
a comprehensive framework for modeling the respiratory, the urinary, the musculo-skeletal, the
human body using computational methods to alimentary, the reproductive, the endocrine, the
provide answers to basic questions, and better haemolymphoid, and the integumental systems.
care for human beings. The collaborative research To illustrate the international collaboration, more
initiatives consist of scientifically independent than 16 research laboratories from five countries
projects on integrative systems physiology and (Australia, USA, United Kingdom, Israel, and

753
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

Switzerland) are currently working only on spatial and temporal scales. Modeling at the sub-
cardio-models. cellular level has advanced to an impressive level
Virtual Physiological Human is a European in most biological tissues, partially guided by
Union initiative which started in 2007. Its main direct knowledge transfer from cardiac to other
targets are the creation of several patient-specific cell models (Youm et al., 2006).
computer models that will be used for personalized Models are formulated in the cellular, intercel-
and predictive healthcare; as well as, the creation lular, tissue, organ, and organism levels. At the
of ICT-based tools for modeling and simulation of organ and organism level, complexity of computa-
human physiology and disease-related processes. tional (and experimental) models increases rapidly.
The Physiome and the Virtual Physiological Hu- In order to handle this problem, the multitude
man projects seek to understand and describe the of interacting processes and components must
human organism, its physiology and pathophysi- be assessed for inclusion into, or elimination
ology, and to use this understanding to improve from, mathematical representation of biological
human health. While it will be a very long time behavior. Different researchers have taken differ-
before a surgery will be executed or a drug’s ef- ent approaches, but applied (i.e., experimentally
fects will be tested on a virtual patient, that day testable) work seems to follow the pattern that,
is closer than ever. But we need to recognize the once the research question has been determined
potential of such international efforts. The most experimentally, the mathematical models are
daunting challenge for the future remains the inte- developed to maximally reproduce relevant be-
gration of this incredible wealth of information to havior with minimal complexity. This process of
increase our awareness of how biological systems selection and reduction is, of course, difficult and
are structured at all levels, and how this structure usually requires a continuous iteration between
drives the function of a healthy or diseased entity. experimental and theoretical model application.
In order to formulate a model description, two
main pathways exist. The first pathway leads to
17.5. LEVELS OF MODELING: a mathematical model via a physical description
FROM CELLULAR TO ORGAN of the system. The second pathway is based on
AND SYSTEMS MODELING the system identification using observations.
These pathways will be further discussed in the
The breadth and depth of the experimental data next Section.
currently obtained across laboratories all over the Starting from the sub-cellular and cellular lev-
world has allowed the design of sub-cellular to els, Hodgkin & Huxley introduced the “paradigm”
whole organ models. For the reasons discussed physiological model of nerve excitation (Hodgkin
in Section 17.4.2, a rapid expansion of detailed & Huxley, 1952). Eight years later, Denis Noble
experimental data, mainly occurred in the last presented the first cardiac cell model (Noble,
decade of the previous millennium, had created the 1960). These two works were the cornerstones
area to develop the “theoretical biology” (Noble, for the development of the current, computational
2002). The term “Systems Biology” represents a approach to modeling of living cells. As we ascend
novel, quantitative approach to biological research the spatial biological ladder, we need to integrate
that encompasses physiological functioning as the cell functioning to a more complicated level of
well. Biology and physiology are merged together structure that resembles that of a tissue. Numerous
using a combination of experimental data and a mathematical and computational descriptions of
quantitative theoretical description of the interac- cellular and inter-cellular effects use the work of
tions between system components across multiple Beeler & Reuter (1977) for models of the electrical

754
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

activity propagation in the intracellular and extra- dynamic model is characterized by a number of
cellular spaces. At the organ and organism level, variables whose values change with time, even
complexity of computational (and experimental) in the absence of external inputs. These variables
models increases rapidly, and scientists usually fully describe the systemic behavior at any given
simplify their models in order to gain insight into time and are known as state variables. On the
the underlying physiological system that is be- other hand, a static model has direct instanta-
ing examined. The work from researchers at the neous links between all variables. A very broad
University of California, San Diego, CA, USA categorization, which is nonetheless quite useful
is a good example of how advanced the field of for creating more finely structured hypotheses,
system modeling has become in this regard. Bigg considers randomness, a priori knowledge of the
is a freely available model of the first complete model’s structure and the domain of description. A
computer model of human metabolism that helps major target in modeling a physiological system
researchers uncover new drug pathways, and un- is to identify these properties through the use of
derstand the molecular basis of cancer and other appropriate computational tools.
diseases (Schellenberger et al., 2010).
17.6.1. Deterministic and Stochastic
Models
17.6. CLASSIFICATION OF MODELS
In a deterministic system, we always have an ex-
We can now begin to get to the heart of the matter, act relationship between measurable and derived
by describing and classifying models. This section variables. Given a clear knowledge of the initial
will deal with models of physiological systems conditions and the system dynamics, the future
and their behavior; either dynamic or static. A behavior of a deterministic system has no uncer-

Figure 3. Classification of models

755
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

tainty for all time. Most physiological systems the sub-processes. These laws, together with physi-
are studied as deterministic and the unavoidable cal constraints, are used to integrate the models of
uncertainty is introduced into the model as a subsystems into an overall mathematical model of
separate random process superimposed into the the system. If one has valid representations from
variables of the system. What makes a determin- basic science, then equations can be postulated to
istic system so desirable is that, given sufficient represent the system under study in either continu-
knowledge about the dynamics and the values of ous time or discrete time (events). In which case,
the state variables at a given time (the state of the the variables are related by equations containing
system at that time), the future course of the system parameters which define system behavior. In the
can be predicted with some degree of accuracy. case of static systems, the relations are simple
On the other hand, the outcome of a stochastic algebraic equations independent of time. For dy-
model is governed by some degree of chance. namic systems (linear or non-linear), the equations
Even if complete information on the dynamics include functions of time and require knowledge
and the initial states of such a model is given, of past values for variables. In addition, the system
the future course of the system is impossible to under study may have lumped variables, or require
be fully predicted. Rather, the model’s output can distributed parameters over a domain of interest
be described in terms of its statistics; that is, the (e.g., temperature in space). The latter is usually
likelihood of its state variable having particular described with partial differential (or difference)
values. A basic modeling question in many ex- equations or finite elements.
perimental situations is whether the system used A nonparametric model provides a method
to provide the acquired data is to be modeled to estimate a system’s output representing the
as deterministic or stochastic. In practice, the actual relationship between the input and the
acquired data set is the result of a mix of deter- output, without making restrictive assumptions
ministic and stochastic processes. In fact, such a about the variables of the system or its statistic
concern is further complicated; we can always properties. Such models can provide accurate
construct a deterministic system that will generate methods of data analysis, because they make
the specific data of any given finite data set, even minimal assumptions about the data-generating
if our data set is acquired from a highly stochastic process. In the nonparametric black box approach,
process. A golden rule for these kinds of situations a mathematical model is formulated on the basis
is this: We always seek to model a process with of the input output characteristic of the system
the maximum possible simplicity. without consideration of the internal function-
ing of the system. Linear nonparametric models
17.6.2. Parametric and consist of data tables representing the impulse
Nonparametric Models response, step response, and frequency response
of the system. Because nonparametric models
For a better description and analysis of any sys- are not represented by a compact mathematical
tem, we need to introduce the subtle distinction formula with adjustable parameters, such models
between variables and parameters. A parameter is do not impose a specific mathematical structure
a constant; it is a term in an equation that is fixed. on the system.
On the contrary, a variable changes with time to Now a question arises on the selection criteria
reflect the dynamics of the system. between those two types of models. The modeling
A parametric model is a bottoms-up representa- choice depends mainly on the nature of the system,
tion of a process based on physical principles and on the type of behavior that is expected, and on the
a-priori knowledge of constitutive laws governing intended use of the model. Nonparametric models

756
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

serve well as preliminary models that are used An even more extensive model would also
to analyze system characteristics. For example, include chain dynamics explicitly, as well
estimating the transient response provides insight as air pressure against the running bicycle.
into the rise time and settling time of the system Knowing which variables are important to
response. Similarly, estimating frequency response include in the model is one of the keys to
might indicate the order of the system, locations successful modeling, and this is, in many
of resonances and notches, crossover frequencies, cases, more an art than a science.
and the bandwidth of the system. In some cases, a Example 17.2: Another example, aimed to distin-
specific mathematical form is preferable because guish between parameters and state variables
the estimated parameters have a physical inter- is given below. In the case of modeling the
pretation. However, when estimates of dynamic heart rhythm during a specific short-term
characteristics are only required, nonparametric physical activity, the subject should not eat
models are usually used. during the exercise, and the exercise should
take place in a limited amount of time so
17.6.3. Applied Examples that circadian fluctuations do not have a
significant effect on the experiment. Hence,
Example 17.1: A simple example of a dynamic food and the time of the day are considered
system is that of a bicycle ride. The state as fixed parameters (i.e., they are constant).
variables of the model include the bicycle’s On the contrary, if we want to model the heart
speed and the feet pressure on the pedals. rhythm over the day, then the time of day
The variables are related in a direct but and food absorption become state variables.
potentially complicated manner. A simple Example 17.3: Any signal that is recorded from
model would just consider speed to be pro- the brain, either inside (e.g., local field
portional to pedal pressure. A more realistic potentials - LFP) or outside of the scalp
model would include time delays resulting (e.g., electroencephalograph - EEG)), is
from the chain dynamics and neural lag. a highly stochastic signal. The LFP is an

Figure 4. A local field potential (LFP) recorded inside the subthalamic nucleus of a Parkinson’s disease
patient. The signal is highly stochastic since it is produced by a stochastic system. The LFPs are domi-
nated by the more sustained currents in the tissue, typical of the somato-dendritic currents.

757
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

electrophysiological signal, dominated by Compartmental modeling is also a significant

slow varying potentials, typical of a neuron’s approach of modeling neural systems. Various
somato-dendritic processes within a volume platforms have been developed to provide the
of tissue. The electrical potential is usually tools for a detailed realistic simulation of a real
recorded with a very small electrode embed- neuron, or even a large network of neurons based
ded within neuronal tissue, typically in the on a “building block” approach. In such systems,
brain of an anesthetized animal or patient simulations are constructed from modules that
(in vivo) or within a thin slice of brain tissue receive inputs, perform calculations on them, and
maintained in a solution (in vitro). A typical then generate outputs. GEneral NEural SImula-
LFP signal, acquired from the subthalamic tion System (GENESIS) is a general purpose
nucleus of a Parkinson’s disease patient, is object-oriented software platform developed by
shown in Figure 4. James Bower and David Beeman (Bower & Bee-
man, 1998) to support the biologically realistic
simulation of neural systems. This object-oriented
17.7. COMPARTMENTAL MODELING environment enables the modification of existing
simulations for new purposes. GENESIS, and
Compartmental modeling is mainly used to its version for parallel and networked computers
describe systems that include transfer of solutes (PGENESIS), was the first broad scale modeling
across compartments, such as the respiratory and system in computational biology to encourage
circulatory systems. It is based on metabolism modelers to develop and share model features and
of tracer-labeled compound studies that started components. It supports the simulation of neural
in the 1920s. Compartmental models are linear, systems, ranging from subcellular components
nonlinear, continuous or discrete models of sys- and biochemical reactions to complex models of
tems that are divided into homogenous well-mixed single neurons, simulations of large networks,
components, called compartments. A compartment and systems-level models.
is a well-delineated biotic or abiotic entity. The An alternative to the GENESIS simulation
models may have constant or even time-varying environment is NEURON (http://www.neuron.
parameters. The internal behavior of the system yale.edu), which is widely used by experimental
is characterized by the movement of materials and theoretical neuroscientists. It was primarily
between two neighboring compartments. Two developed by Michael Hines, John W. Moore, and
of the main difficulties of compartment model- Ted Carnevale at Yale University, New Haven,
ing are the determination of the exact number of CT, USA and Duke University, Durham, NC,
compartments to be used in the model, and the USA (Hines & Carnevale, 1997). Both platforms
accessibility of some of the compartment’s data. implement a built in “scalability” in models. This
Lumped compartmental variables are mainly sub- is a major advantage compared to other custom
stances (solutes) that are either exogenous (e.g., made codes needed to be written for a specific
a drug) or endogenous (e.g., insulin). Blood and simulation (e.g., in a MATLAB & Simulink en-
chemical species (such as hormones) distribution vironment), but it comes with the expense of a
to various organs, cellular dynamics, tempera- need to invest the time required to understand the
ture distribution, etc. are just few examples in analysis and graphic tools provided by platforms
which compartmental models are used in studies such as GENESIS and NEURON.
involving pharmacokinetics, chemical reaction
engineering, fluid transport etc.

758
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

17.7.1. Detailed 17.7.2. Modified

Compartmental Models Compartmental Models

In order to describe the transfer of a solute by dif- The compartment analysis presented in Section
fusion between two compartments, the following 17.7.1 is not adequate to fully describe systems
assumptions are needed: in which the transfer rates are not constant, but
depend, for example, on the concentration of a
1. All compartments have constant volumes. solute in a single compartment. But even in those
2. The solutes, upon entering a compartment, systems, we can apply a modified compartment
are dispersed homogenously in the entire analysis to cope with the nonlinearities present.
compartment. As the model becomes more and more complex,
3. The rate of solute depletion from a compart- an analytical solution is not feasible; yet, simula-
ment is analogous to the concentration of the tions of such models can give us an approximation
solute in the same compartment. of the solution.
One of the earliest modeling attempts that
If the aforementioned assumptions are met, aimed at analyzing smallpox morbidity and mor-
the time course of a solute transfer across two tality dates back to 1766 when the Dutch-Switch
compartments can be examined. Using a law of mathematician, Daniel Bernoulli, tried to analyze
diffusion derived by Adolph Fick in the year 1855, it as a statistical problem to demonstrate the ef-
we can model the diffusion coefficient, D of a ficacy of vaccination. The next infectious disease
solute, transferred between two compartments, modeling attempt belongs to Hammer and Soper
that has quantity, q, and concentration, c, using who created a model of measles spreading, in
a membrane with surface area, A, and thickness 1906. Their model contained separate compart-
dx, as follows, ments for susceptibles, infectives, and recovered,
taking into consideration the births, the infection
dq dc rate, etc. Twenty years later, Kermack and McK-
= −DA (17.1)
dt dx endrick (in the continuous time), and Reed and
Frost (in the discrete time) presented extensions
The transfer rate, R, of the diffusion is defined as for the model of Hammer and Soper.
For both the Kermack-McKendrick and Reed-
Frost models, any given person is related to a
DA
R= (17.2) certain time period. The latent period is the time
dx
elapsed between contact and the actual discharge
of the infectious agent. The infectious period is
For a thorough review and an analytical ap- the time during which the contagious agent is
proach of two-compartment models, please see spread to others. The immune period is the time
(Enderle, 2005). The simplification of compart- during which a person has temporal or permanent
ment models is allowed by the fact that the dis- immunity and can no longer transmit the agent.
tribution inside a compartment is not included. The incubation period is the time elapsed between
The basic assumption of a solute homogeneously contact and the observation of symptoms. The
mixed inside a compartment, results in knowing symptomatic period is the time interval in which
everything about a system’s behavior, when the the person overtly displays signs of the illness; see
inflow and outflow for each compartment are (Enderle, 2005) for an illustration of these periods.
identified.

759
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

If we consider a population of size, n, with x and then become permanently immune; v) since
susceptibles, y infectives, and z immunes, so that the person can be infected at any instant during the
n = S + C + R, the assumptions for a Kermack- time period, the average latent period is one-half
McKendrick continuous time modeling approach of the time period, where the length of the time
are the existence of: i) a uniform mixing among period represents the period of infectivity; and, vi)
the population, ii) a zero latent period, iii) a closed each individual has a fixed probability of coming
and isolated population, iv) a negative exponen- into adequate contact p with any other specified
tial distribution for the infectious period, v) an individual within one time period.
infectious rate, b, and vi) a removal rate, g. The The structure of the Reed-Frost model is shown
course of an infectious epidemic in a closed and in Figure 5 (bottom panel). Note that the prob-
isolated population is a function of the number ability of adequate contact p can be thought of as
of susceptibles and the infectious rate between
susceptibles and infectives. In Figure 5 (upper n
panel), the Kermack-McKendrick model is shown. p= (17.4)
N
Arrows indicate a nonnegative transfer of indi-
viduals from one state to another, dependent on
the infective rate b (infectives) and the removal where, n is the average number of adequate
rate g. The Kermack-McKendrick model describes contacts.
the transfer of S susceptibles, C infectives, and R As before, the Reed-Frost model describes
immunes at time t from state to state. With b as the transfer of S susceptibles, C infectives, and R
the infective rate, the differential equations that immunes, but now the transfer is measured with
describe the model are: respect to the next discrete time (state), k+1. After
adequate contact with an infective in a given time
dS period, a susceptible will develop the infection and
= −bSC be infectious to others only during the subsequent
dt
dC time period, after which one becomes immune.
= bSC − gC (17.3) If the infective rate is (1 - qC(k)), the model can be
dt
dR described by the nonlinear difference equations
= gC
dt
C (k + 1) = S (k )(1 − q C (k ) )
Equation 17.3 can be solved analytically using S (k + 1) = S (k ) − C (k + 1) (17.5)
a Taylor series expansion; see (Enderle, 2005). R(k + 1) = R(k ) + C (k )
The Reed-Frost model is a deterministic dis-
crete time model; this makes it more practical in
The time period T is understood to be the
being used with true data which is usually sampled
length of time an individual is infectious, so that
versions of continuous data. The assumptions for
the removal rate is equal to one.
a Reed-Frost discrete time modeling approach
are as follows: i) the existence of a uniform
17.7.3. Expansion to Multi-
mixing among the population; ii) the existence
Compartmental Models
of a zero latent period (although the model can
extend easily to a nonzero latent period having a
It should be clear by now that real biological
well defined distribution); iii) the existence of a
models incorporate more than the limited number
closed population at steady state; iv) susceptible
of compartments already described in previous
individuals can develop the infection only once
sections. A single compartment model can be

760
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

Figure 5. Modified compartmental models. The Kermack-McKendrick (upper panel) and the Reed-Frost
(bottom panel) models are shown.

divided to multiple compartments if we choose organs or tissues, and their interconnections cor-
to include more details on it such as cell volume, respond to blood or lymph flows. This modeling
interstitial volume, or plasma volume. But even approach aims to balance between complexity
these volumes can be further compartmental- and simplicity to predict the absorption, dis-
ized. For instance, the interstitial volume can tribution, metabolism and excretion (ADME)
be defined with compartments including the GI of synthetic or natural chemical substances in
tract, mouth, liver, kidneys, and other unidentified humans and animal models. PBPK models may
compartments. Each of these compartments has its have purely predictive uses, but other uses, such
own transfer rate for moving the solute from one as statistical inference, have been made possible
compartment to another. In general, concern about by the development of various statistical tools. A
how the solute moves from and into a compart- system of differential equations for concentration
ment is not a focus, but only the amount of solute or quantity of substance on each compartment is
that is transferred. The concepts described in the usually written, and its parameters represent blood
previous section can be applied to a model with flows, pulmonary ventilation rate, organ volumes,
any number of compartments. Each compartment etc. PBPK models are also used for inter-species
is characterized by a conservation of mass differ- transpositions or extrapolation from one mode of
ential equation that describes the rate of change administration to another (e.g., males to females,
of solute. Thus, for the case of N compartments, adults to infants, inhalation to oral) to asses toxicity
there are N equations of the general form risk and therapeutic drug development.

dqi 17.7.4. Applied Example

= Input − Output (17.6)
dt
Example 17.4: Let us consider a two compart-
ment model, shown in Figure 6, in analogy to the
where, qi is the quantity of solute in compartment i.
one presented by Goodman and Noble (1968).
For a linear system, the transfer rates are constants.
According to that model, the rate of cholesterol
Physiologically based pharmacokinetic
turnover has been described as conforming to a
(PBPK) modeling is a multi-compartmental
two-compartmental system consisting of one pool
modeling technique used in pharmaceutical re-
that turns over rapidly and a second pool with a low
search and drug development, and in health risk
turnover rate. Cholesterol is inserted into the blood
assessment for cosmetics or general chemicals.
plasma of all animals by two sources, namely the
Compartments correspond to a-priori defined

761
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

food and synthesis from simpler substances within liver compartment. The cholesterol that is inserted
the body. Cholesterol is recycled. It is excreted by via food and from biosynthesis is not radioactive.
the liver via the bile into the digestive tract. This Hence, the only inflow of cholesterol-C14 into
system can be described using a two-compartment the first compartment arrives from the second
model, where some of the tissue (primarily the compartment. Now, let us consider the second
liver, which is the main organ playing a role in compartment. The total inflow of cholesterol into
the dynamics of the cholesterol levels) and the the second compartment is R2+R3 and the amount of
blood exchange cholesterol with the blood. We the cholesterol-C14 is C2(t)=Q2(t)/G2. This means
assume that the exchange of cholesterol between that the amount of cholesterol-C14 that flows into
the liver and the blood is happening in a high, the first compartment is (R2+R3)Q2(t)/G2, and the
almost instantaneous speed; that is why we model amount of cholesterol-C14 that flows out of the
the blood-liver system as a single compartment. first compartment and into the second compart-
The rest of the exchange – between blood and ment is R3Q1(t)/G1. The amount of cholesterol-C14
liver and the rest of the tissues, lumped together that is extracted to the environment is given by
in a second compartment – is happening at a (R0+R1+R2)Q1(t)/G1. From these relations, we can
much slower speed. Hence, the first compart- write down the differential equations that govern
ment represents the amount of cholesterol in the the system as follows,
blood and liver, and the second represents the
amount of cholesterol in all the rest of the body. R2 + R3 R0 + R1 + R2 + R3
Q1' (t ) = Q2 (t ) − Q1 (t )
If we inject a small amount of C14 into the blood G2 G1
stream, we can estimate the amount of radioactive R3 R2 + R3
Q2' (t ) = Q1 (t ) − Q2 (t )
cholesterol in the two compartments, Q1 and Q2 G1 G2
respectively. We assume that the concentration of (17.7)
radioactive cholesterol-C14 in the first compart-
ment is C1(t)=Q1(t)/G1, and the concentration of
radioactive cholesterol in the second compartment
is C2(t)=Q2(t)/G2. Let us first consider the blood-

Figure 6. Compartmental model of cholesterol concentration in the body

762
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

17.8. LINEAR MODELING OF mation, etc.) will affect a change on the output
PHYSIOLOGICAL SYSTEMS in exactly the same way as if the transformation
were applied to it directly. That is, if f and g are
Linear systems are highly popular among the two linear operators, then f(g(x))=g(f(x)).Thus, for
physiological models since they are simple to example, the response of a linear system to a step
implement and provide extremely powerful tools input can be computed by integrating its impulse
for their analysis. In contrast, methods available response, since a step is the integral of an impulse.
for the study of nonlinear systems are much more Under the same test conditions, a system that
limited. In fact, almost all physiological systems is time-invariant will respond identically to a
are nonlinear; however, many of these systems specific stimulus irrespective of when it is intro-
can be modeled as linear systems in a limited duced. That is, except for the time shifts between
range of operation. responses, all responses are identical. Just as not
Let us introduce a short description of the all systems are linear, not all linear systems are
terminology in the field. If the operation that time-invariant. Mathematically, time invariance
transforms the input into the output varies with can be expressed as follows:
time, the system is time varying; whereas, if the
operation remains constant, the system is time y(t ) = f (x (t )) ⇒ y(t − t ) = f (x (t − t ))
invariant. Two attributes of linear time-invariant (17.9)
(LTI) systems form the basis for almost all analyti-
cal techniques applied to these systems: where, τ is a time constant. A system that satisfies
both of these properties is naturally called a linear
1. Response obeys the principle of superposition. time-invariant (LTI) system.
2. Response can be expressed as the convo- Testing a system for linearity may be done
lution of the input with the unit impulse using the principle of superposition. An easy way
response of the system. to implement such tests is to apply the same input
at different amplitudes. If the system is linear, the
The concepts of superposition, convolution, output will have the same shape and the output
and impulse response will now be defined shortly. amplitude will scale with the input amplitude.
The principle of superposition states that if the It is also useful to remember that the response
system has an input that can be expressed as a of a linear system to a sinusoidal input will be a
sum of signals, then the response of the system sinusoid at the same frequency. Thus, if the output
can be expressed as the same sum of the individual has components at frequencies not in the input, it
responses to the respective signals. Superposition must be nonlinear.
can be expressed mathematically as follows: Many systems behave linearly over a restricted
range of inputs. For example, a rectifier is linear
f (a1x 1 + a2x 2 ) = a1 f (x 1 ) + a2 f (x 2 ) (17.8) as long as the input remains either positive or
negative. Almost any system will become non-
linear if the input is large enough. Conversely,
where, x1 and x2 are two inputs, f(x1), f(x2) are
most nonlinear systems can be described by a
the respective outputs of a system, fand a1, a2 are
linear approximation if the input amplitude is
two scalars. Superposition applies if and only if
small enough. Thus, it is important to determine
a system is linear. The effects of performing any
not only whether a system is linear, but over what
linear operation on the input of a linear system
range of values does it behave linearly. Thus, it is
(e.g., integration, differentiation, Fourier transfor-
important to determine the linear range of a system.

763
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

In some cases, a system may have more than one One way to characterize the dynamic behavior
linear range and display different behaviors in of a linear system is in terms of its response to
each range (e.g., a full wave rectifier). Note that an impulse. The impulse response function (IRF)
the linear range is a property of the amplitude of can be used as a representation of a linear system
the input – not its frequency content. That being because it can be used to predict the response of
said, the linear range may vary with frequency the system to any input. To visualize how this
for some types of nonlinear systems. works, consider the input to the system to be a
An approach that is frequently useful in dealing series of impulses of different amplitudes. The
with nonlinear systems is to transform either the response of the system to any one impulse is
input or the output in order to make the resulting simply the IRF multiplied by the amplitude of the
input-output relation more linear. For example, input impulse, and delayed by the time at which
logarithmic transformations are useful in linear- the input impulse occurs. Now, because a linear
izing systems in which there is a power relation system obeys the superposition principle, the
between input and output. overall output is simply the sum of the responses
to all the input impulses. The convolution integral
17.8.1. Time-Domain and is the mathematical statement of this procedure.
Frequency-Domain Models A system’s IRF can have both positive time
values, representing system memory, and negative
If a system is known to be linear, it is always time values, representing system anticipation. The
guaranteed that an adequate model of the system response, y(t), of such a two-sided IRF, h(t), to an
can be determined. This consists of determining input, x(t), is given by the convolution integral
the system’s response to a set of basis functions
(for example impulses or sinusoids of different ∞

frequencies). Once these responses are known, the

y(t ) = ∫
−∞
h(t )x (t − t )d t (17.10)
response to an arbitrary input may be determined
as follows: ∞
If, as is usually the case, ∫ h 2 (t )d t < ∞ ,
−∞

1. Decompose the arbitrary input into a linear then the system has finite memory and h(t ) @ 0
combination of basis functions (e.g., Fourier when t < T1 and t > T2 for some value of T1
analysis decomposes the signal into a linear and T2. Under these conditions, Equation 17.10
combination of sinusoids). may be simplified to
2. Determine the response to each component
using the principle of proportionality. T2

3. Sum the resulting components to determine

y(t ) = ∫
T1
h(t )x (t − t )d t (17.11)
the overall response by relying on the prin-
ciple of superposition. In causal (physically realizable or non-antici-
patory) systems there is no anticipatory component
If the basis functions are a series of impulses, to the response; e.g., h(𝜏)=0 for τ<0 so that T1=0.
then the analysis results in time domain models. The IRF is then one-sided and the convolution
On the other hand, if the basis functions are integral further simplifies to
sinusoids, then the analysis results in frequency
domain models. T2
y(t ) = ∫ 0
h(t )x (t − t )d t (17.12)

764
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

A linear system can be represented by either a estimates of the system’s dynamics. However, the
parametric or non-parametric IRF. A parametric presence of a feedback relation can be detected
IRF is in the form of an equation. The structure as an anticipatory component of the IRF, relating
of the equation defines the class of systems it the input to the output. Hence, computing the
represents, and the parameters of the equation two-sided IRF provides a means of testing for a
determine how the behavior differs from that of feedback relation between two signals.
the other members of the same class. In contrast, a Now, we will deal with an alternative approach
non-parametric IRF consists of the sampled values in which linear dynamics are characterized in terms
of the response, and is stored as a real vector in of the response to sinusoidal stimuli of different
the time domain. In short, the parametric form frequencies. The response of a linear system to
can be represented by an equation, and the non- a sinusoidal stimulus will be a sinusoid of the
parametric form can be represented by a curve. same frequency but of different amplitude and
It is normally assumed that physical systems phase. The frequency response of a linear system
are causal and do not anticipate. Consequently, describes the relative magnitudes of the input and
the usual IRF identification procedures employed output sinusoids (gain), and the phase difference
in engineering determine only the positive or as a function of frequency.
memory part of the IRF. There are a number of The frequency response of a linear system may
areas of research, particularly those involving the be used to determine the response of the system
life sciences, where it is important to determine to an arbitrary input as follows:
both the positive and negative parts of an IRF.
Two-sided IRFs will be important in situations 1. Decompose the input signal into a sum of
involving actual prediction. Living systems sinusoids using Fourier analysis.
frequently demonstrate predictive behavior. For 2. Multiply each sinusoid by the gain of the
example, the frequency response of the visual system at the appropriate frequency, and
pursuit system is wider for predictable stimuli shift it by the corresponding phase.
than for random stimuli. Effective prediction can 3. Sum the scaled and phase-shifted sinusoids
occur when the input is unknown but structured to reconstruct the overall response.
(e.g., periodic), or when preview of the input is
possible. Under such conditions, a negative por- The response of a linear system to an arbitrary
tion of the IRF may well occur. A pure delay of input may be computed from its impulse response
𝜏, either preceding or following a linear dynamic using the convolution integral defined in Equa-
system, moves the IRF 𝜏 to the right. Thus, whether tion 17.10. The Laplace transform of this relation
or not a system contains a pure delay may often gives Y(s)=H(s)X(s); where, H(s) is the Laplace
be determined from the IRF. If the input to the transform of the impulse response, Y(s), X(s) are
system is measured after a delay of 𝜏, then the IRF the Laplace transforms of the output and the input,
is shifted to the left with the result that negative respectively, and s is a complex variable defined as
time values may occur, necessitating the use of s = σ + jω, σ being a damping factor and ωbeing
two-sided IRF identification techniques. Once a frequency term. The transfer function of the
the delay has been determined from the identified system can then be written as
IRF, the input can simply be shifted with respect to
the output to eliminate the delay. There are many Y (s )
H (s ) = (17.13)
situations where the input to a system is related to X (s )
its output by feedback. Attempting to identify the
system under such conditions can lead to incorrect

765
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

The transfer function of any linear, time-in- AH (−j w) AH ( j w)

c1 = , c2 = (17.20)
variant, constant-parameter system without delays 2 2
may be written as the ratio of two polynomials:
so that
(s − z 1 )...(s − z m )
H (s ) = K (17.14)
(s − p1 )...(s − pm ) (
yss(t) = A H ( j ω) cos ωt + φ H ( j ω) )
(17.21)
where, the zeros (zi) and poles (pi) of the polyno-
mials may be real, zero, or complex (if complex where, |H(j𝜔)| denotes the magnitude of H(j𝜔),
they come as conjugate pairs). To determine the and φ H ( j ω) is its phase.
frequency response of a system with transfer func- Thus, the steady state sinusoidal response of
tion H(s), apply a sine wave stimulus: a linear system can be operationally determined
from its transfer function by letting s=j𝜔, and then
x (t ) = A sin(wt ) (17.15) evaluating the magnitude and phase of the result-
ing complex number as a function of frequency.
which has the Laplace transform X (s ) = Α w
. Conversely, the frequency response of a system
s 2 + w2 can often be used to determine the underlying
The response in the Laplace domain will be transfer function.
Sinusoidal inputs provide a convenient,
s (s − z 1 )...(s − z m ) straightforward means of determining the fre-
Y (s ) = AK
(s 2 + w 2 ) (s − p1 )...(s − pn ) quency response of a system. The procedure is
(17.16) as follows:

Expanding the right hand side of Equation 1. Apply a sinusoidal stimulus at frequency
(17.16) using partial fractions gives ω to the system, wait for the response to
reach steady state, and record the resulting
c1 c2 c3 c4 sinusoidal response.
Y (s ) = + + + + ... 2. Compute the ratio of the response amplitude
s + jw s − jw s − p1 s − p2
(17.17) to the input amplitude, and use it as a measure
of the system gain at frequency ω.
Taking the inverse transform gives the solution 3. Compute the phase shift of the output with
respect to the input, and use it as a measure
p1t p2t of the system phase shift at ω.
y(t ) = c1e − j wt + c2e j wt + c3e + c4e + ...
4. Repeat steps i-iii at frequencies over the
(17.18)
range for which the system responds.
5. Draw or fit a smooth curve through the
All pi, i=1,…,n must be less than zero for the
resulting points.
system to be stable, so the steady state response is
Advantages of sinusoidal testing include:
yss (t ) = c1e − j wt + c2e j wt (17.19)
1. The gain of the recording system can be
Standard partial fraction techniques then give adjusted at each frequency (either manually

766
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

or automatically) to use the full dynamic a stationary, random input x(t) which generates the
range and minimize the effects of noise. stationary random process y(t) as output. Then,
2. The amplitudes of the input sinusoids can be
adjusted until the output amplitude reaches t

some desired value.

y(t ) = ∫ 0
h(t )x (t − t )d t (17.22)
3. In the presence of noise and nonlinearities,
only the amplitude and phase of the sinusoi- The autocorrelation function of the output is
dal component at the input frequency need given by
be measured.
Ryy (t ) = Ε y(t )y(t − t ) (17.23)
Sinusoidal testing is very effective, when
practical, but does have a number of limitations:
which has the expected value
1. The approach requires the application of pure
t t
sinusoids of a single frequency. This is often
difficult technically. Furthermore, in the life
∫ ∫0 0
h(v )h(m)Rxx (t − m + v )dvd m
(17.24)
sciences, particularly in behavioral studies,
it is often desirable to avoid predictable,
where, Rxx is the autocorrelation of the input. Thus,
periodic stimuli.
the output autocorrelation function is defined by
2. The procedure is time consuming. Each
the system’s impulse response and the autocorrela-
stimulus frequency must be applied sepa-
tion function of the input. The cross-correlation
rately and the response recorded only after
function Rxy between the input x(t) and the output
the transient response has decayed. If the
y(t) may be derived from the relation
system’s time constant is long, this may
require many cycles at each frequency.  ∞ 
The time taken to do sinusoidal testing is Ε x (t )y(t + t ) = Ε  ∫ (h(v )x (t )x (t + t − v )dv 
 0 
particularly important in the study of physi- (17.25)
ological systems where experimental time is
always limited. In addition, living systems which has the expected value
are frequently time varying so it is important
to obtain an identification in as short a time ∞
as possible. Rxy (t ) = ∫ 0
h(v )Rxx (t − v )dv (17.26)
3. Only a limited number of frequencies can
be tested. If too few frequencies are tested,
Thus, the cross-correlation between the input
sharp changes in the frequency response,
and output is simply the convolution of the input
e.g. resonances, may be missed.
auto-correlation function, with the Fourier that is
transforming these relations, yields the frequency
17.8.2. Stochastic Testing domain expressions:

Consider a constant parameter, linear system de- 2

scribed by the one-sided, impulse response h(𝜏) Syy ( j w) = H ( j w) S xx ( j w) (17.27)
with the corresponding frequency response func-
tion H(j𝜔). Assume that the system is subjected to and

767
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

S xy ( j w) = H ( j w)S xx ( j w) (17.28) The coherence-squared function will have

values in the range 0 to 1, and is analogous to the
variance accounted for as a function of frequency
where, Sxx(j𝜔)and Syy(j𝜔) are the input and the out-
(i.e., the square of the correlation coefficient
put power spectra, and Sxy(j𝜔) is the input-output
function which arises in linear regression). For a
cross spectrum. The gain portion of the system
constant parameter linear system with no noise,
frequency response may be estimated from the
the coherence-squared will identically equal to 1.
input and output power spectra as
If the input and output are completely unrelated,
the coherence-squared function will have a value
2 Syy ( j w)
H ( j w) = (17.29) of 0. If the coherence-squared function is greater
S xx ( j w) than zero but less than one, three possibilities exist:

However, this estimate gives no information 1. Extraneous noise is present in the

about the phase. Moreover, it will be biased if measurements.
there is noise at either the input or the output. A 2. The system is not linear.
better approach is to determine the system fre- 3. y(t) is an output due to an input x(t) as well
quency response function from the input power as to other inputs.
spectrum and the input-output cross spectrum by
using Equation 17.26 to get the relation: The coherence-squared can be interpreted as
the fraction of the output variance that is linearly
S xy ( j w) related to the input at each frequency.
H ( j w) = (17.30) Note that the coherence function is usually
S xx ( j w)
estimated from spectral estimates obtained by
averaging a number of segments of the original
Sxyis a complex number, so the frequency re- data. The bias error associated with coherence
sponse has both a magnitude (or gain) and a phase estimates varies with the number of segments
characteristic. Moreover, because of the averaging and the expected value of the coherence; the error
involved in computing the cross-spectrum, the decreases as either or both increase. Estimates of
estimate will not be biased as a result of output the coherence function may be in serious error
noise. However, if there is much output noise, if the number of segments is small and/or if the
then long data records and, hence, much averag- value of the coherence function is low. Indeed,
ing may be needed to reduce the random error. the worst case occurs if only one segment is
Furthermore, noise at the input will still result in used to estimate the coherence function, since
biased results. the coherence estimate will always be equal to
The coherence squared function between the one for this case.
input x(t) and the output y(t) of a system is a real- The procedure for doing frequency analysis
valued function defined by: of a system using a stochastic input is:

S xy ( j ω)
2
1. Apply a stochastic input having power over
2
γ ( j ω) =
xy
(17.31) the range of frequencies where the system
S xx ( j ω)Syy ( j ω) is expected to respond.
2. Record the input and resulting output.

768
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

3. Compute the input spectrum, the output spec- and the procedure to get the differential equa-
trum, and the input-output cross spectrum. tion from the graphic description, which is
4. Evaluate the gain, phase and coherence using based on the Kirchhoff’s circuit laws.
Equations 17.29 through 17.31.
Let us consider the simplest dynamic linear model
Note that since the stochastic input has (“leaky integrator”) of a nerve cell, depicted in
power over a wide range of frequencies, the Figure 7. The resistors R1, R2, R3 represent the
stochastic technique can be thought of as testing neuron’s dendrites and the respective voltages
a large number of frequencies simultaneously. V1, V2, V3 are generated by the synapses from
Consequently, it takes much less time than pure other neurons. The respective currents, I1, I2,
sinusoidal testing. Furthermore, the coherence I3 are integrated in the capacitor C that models
provides a quantitative measure of how well the the cell body membrane capacity. The presence
resulting linear model describes the system. If the of the membrane resistance, R4 denotes that the
coherence function is less than one, it is useful to integrator is “leaky”. The differential equation of
determine whether this is due to additive noise or this model can be written as follows:
due to nonlinearities. One way to investigate this
is to increase the amplitude of the input signal; if 1. We regard all the currents to have a direction
the problem is noise, then the coherence function towards out of the point, VC.
should increase since the output signal-to-noise 2. Kirchhoff’s current law says that the sum
ratio (SNR) should increase. Conversely, if the of all currents in a single node is equal to
problem is nonlinearity, then the coherence func- zero, I R + I R + I R + I R + IC = 0
tion will stay the same or will decrease. Another
1 2 3 4

3. We replace the currents by their voltage

possibility is to repeat the experiment a number of
values as follows:
times with the same input, and average the input VC −V1 VC −V2 VC −V3 VC dVC
and output signals before doing the analysis. If + + + +C =0
R1 R2 R3 R4 dt
noise is the problem, then the coherence of the 4. We can write the same equation in a
average signals will be greater than that of the simplified form,
individual trials. If the problem is nonlinearity, 1   
 + 1 + 1 + 1 V − V1 + V2 + V3  + C dVC = 0
then the results will not change.  R
 1 R2
R3
R 4
 C 
 R
 1 R2
R3

dt

17.8.3. Applied Examples This is a standard form for a first order dif-
ferential equation that can be solved analytically
Example 17.5: A method to model a dynamic or numerically, or even be transformed to the
linear physical system uses simple basic Laplace domain for further analysis.
electrical components, namely a resistor (R),
an inductor (L), a capacitor (C), and sources Example 17.6: A class of simple, yet accurate
of potential (V) and current (I). Such method models, estimated from microelectrode
allows for a more natural modeling approach, recordings, can predict spike generation of
since the system has a direct correspondence single and multiple subthalamic nucleus
to its graphic representation that is more (STN) neurons of Parkinson’s disease (PD)
comprehensible than differential equations. patients. The most characteristic attribute
In this example, we will introduce the basic of an STN neural recording is the presence
elements of an electrical model of a system of bursting/quietness segments. It has been

769
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

Figure 7. Associative linear neural network

suggested that the STN sends the burst- it misses one or two spikes (per 50 ms bin),
ing pulse of spikes as a braking signal to explaining the small jittering observed in
reset the major basal ganglia output nuclei. the exact spike prediction.
This mechanism does not work right in
abnormal situations, such as the PD. To
be able to have a quantitative validation of
the prediction of the model, the coherence 17.9. NONLINEAR MODELING OF
between the predicted spike rhythm and the PHYSIOLOGICAL CONTROL
recorded one is estimated. In Figure 8, the SYSTEMS
two rhythms, calculated in 50 ms bin, and
their coherence are shown. The coherence Any system which violates the principle of super-
is 1 in low frequencies and drops after 2 Hz position is non-linear. Many physical and virtually
(3 dB point is calculated at 2.4 Hz). This all biological systems are nonlinear. In this case, it
depicts that the model predicts the ups and is impossible to provide a general system descrip-
downs of the rhythm accurately; whereas, tion that can be used for any input, and applied at

Figure 8. Coherence estimate between the predicted and the observed spiking rhythm. The prediction is
done using a model that accepts the local field potentials as its input.

770
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

any time. Instead, functional series are typically ∞

used. A functional is a function whose argument y(t ) = k 0 + ∫

0
k1(τ )δ(t − τ )d τ
∞ ∞
is a function and whose value is a number. For +∫ ∫ k (τ , τ )δ(t − τ )δ(t − τ )d τ d τ
0 0 2 1 2 1 2 1 2

instance, the convolution integral evaluated at a

∞ ∞ ∞
+∫ ∫ ∫ k (τ , τ , τ )δ(t − τ )δ(t − τ )δ(t − τ )d τ d τ d τ
3 1 2 3 1 2 3 1 2 3
0 0 0

given time is a functional. +...

17.9.1. Volterra Series (17.33)

An example of a functional series to describe a or equivalently,

non-linear system is the Volterra series. Volterra
showed that if a system is time invariant, has finite y(t ) = k 0 + k1 (t ) + k2 (t, t ) + k 3 (t, t, t ) + ...kn (t,..., t ) + ...
memory, and is analytic (differentiable), then the (17.34)
relation between input x(t) and output y(t) can be
expressed as the infinite sum Hence, the use of impulses to isolate kernels
of different order is not practical here. Another
y(t ) = k 0 + ∫
∞
k1(t )x (t − t )d t problem is that full description of a non-linear
0

+∫
∞

∫
∞
k2 (t1, t2 )x (t − t1 )x (t − t2 )d t1d t2 system with Volterra series, theoretically, has an
infinite number of terms. Because the importance
0 0
∞ ∞ ∞
+∫ ∫ ∫ k 3 (t1, t2 , t 3 )x (t − t1 )x (t − t2 )x (t − t 3 )d t1d t2d t 3
+...
0 0 0
of each functional depends on the form of the
(17.32) non-linearity, and because the terms in this series
are not orthogonal to each other, then
where, k0, k1(τ), k2(τ1,τ2), k3(τ1,τ2,τ3), … are the
kernels of the system, and are symmetric functions 1. One cannot know a priori when or where
of their arguments. The zero-eth order kernel k0, a to truncate the series (small kernels can be
constant, can be assumed to be zero without loss of followed by an important large kernel at
generality by assuming y(t) = 0 when x(t) = 0 (in higher dimensions).
other words, we remove the non-zero bias). The 2. Adding terms changes all the previously eval-
nth order kernel describes the pattern of interac- uated kernels and they must be recomputed.
tion between n pieces of the past stimulus and its
contribution to the total response. However, there 17.9.2. Wiener Series
are other contributions due to nth order interac-
tions also present in all other terms with kernels To address the above issues, Wiener proposed a
of order greater than n. That is the response to special form for a functional series description
nth-order interactions is defined by all kernels of a non-linear system. Assuming white Gauss-
of order n or greater; it is not isolated in the nth ian noise as the input, the Gi functionals in the
component. For example, the first-order term in series are designed to be orthogonal with respect
the series is exactly the same as the convolution to each other and with respect to white noise
integral in a linear system, where the first kernel input functionals of lower order. As a result, the
then represents the impulse response. However, importance of Wiener functionals in the series
note that in a non-linear system, the above series usually decrease in magnitude with kernel order,
defines the impulse response (1st order effect) to and adding terms does not affect already computed
depend on all kernels functionals. Furthermore, the mean squared error
associated with truncation of the series is lowest

771
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

for Wiener descriptions, when compared to other

series truncated at the same order (like Volterra). ∞
2m ! P m −n
Starting with the general Volterra series form, h2n (σ1,..., σ2n ) = ∑ 2n !(m − n )! 2
m =n
m −n
×

Wiener proposed ×∫ ...∫

∞ ∞
k2m (τ1, τ1,..., τm −n , σ1,..., σ2n )d τ1 ...d τm −n
0 0
∞
m + 1)! P m −n
(2m
h2n +1(σ1,..., σ2n +1 ) = ∑ m −n
×
∞ m =n (2n + 1)!(m − n )! 2

y(t ) = ∑ Gm hm (t1, t2 ,...tm ); x (t '), t ' ≤ t 

∞ ∞
×∫ ...∫ k2m +1 (τ1, τ1,..., τm −n , σ1,..., σ2n +1 )d τ1 ...d τm −n
0 0
m =0
(17.35)
(17.38)
where, Gm are now orthogonal functions, x(t) is a
Gaussian white-noise signal with zero mean, and This makes it clear that Wiener kernels, in
hm is the set of Wiener kernels. Each hm is a sym- contrast to Volterra kernels, are polynomial func-
metrical function with respect to its arguments. tions of P, the power of this noisy stimulus. Also,
The first four Wiener kernels are defined by the a given Wiener kernel is a function of higher order
following functionals: Volterra kernels.

G 0 h0 ; x (t ) = h0 17.9.3 Applied Example

∞
G1 h1 ; x (t ) = ∫ h (t )x (t − t )d t
1
0

Example 17.7: A special class of Volterra-Wiener

∞ ∞ ∞
G2 h2 ; x (t ) = ∫ ∫ h (t , t )x (t − t )x (t − t )d t d t − P ∫ h (t , t )d t
2 1 2 1 2 1 2 2 1 2 1
0 0 0

non-linear models is the block oriented

∞ ∞ ∞
G 3 h3 ; x (t ) = ∫ ∫ ∫ h (t , t , t )x (t − t )x (t − t )x (t − t )d t d t d t
3 1 2 3 1 2 3 1 2 3
0 0 0

non-linear models in which a linear time

∞ ∞
−3P ∫ ∫ h (t , t , t )x (t − t )d t d t
3 1 2 2 1 1 2
0 0

(17.36) invariant (LTI) dynamic block is preceded

and/or followed by a static non-linearity.
where, x(t) is Gaussian white noise of zero mean When the linear dynamic block is preceded
and power density spectrum φxx(f)=P (or other- by a static input non-linearity, the model is
wise, autocorrelation ϕxx (τ ) = P δ(τ ) . The func- referred to as a Hammerstein model; and,
tionals have been selected to be orthogonal to when the linear dynamic block is followed
each other so that by a static output non-linearity, the model
is referred to as a Wiener model. Both are
{
E Gi hi ; x (t ) G j h j ; x (t ) = 0 for all i≠j
  } a special case of the situation in which the
linear dynamic block is sandwiched between
(17.37) two static non-linear blocks, a Hammerstein-
Wiener (H-W) model.
Furthermore, Wiener constructed the function-
als so that a given Gk is orthogonal to all homog-
enous functionals of x(t) whose order is less than Briefly, in state space, an H-W model is repre-
k, when x is white noise. For example, if x(t−τ) sented by
is an homogenous functional of order 1, then
E {Gk [hk ; x (t )](t − t )} = 0 , for k > 1. The kernels
v (k | k ) = f (u(k | k ))
in a Volterra series {k} can be related to those in x (k + 1 | k ) = Ax (k | k ) + Bv(k | k )
a Wiener series {h} according to even or odd LTI = 
 w(k | k ) = Cx (k | k ) + Dv(k | k )
terms: 
y(k | k ) = h (w(k | k ))
(17.39)

772
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

where, u∈ℜ is the physical input to the plant, model, and then estimate the parameter values.
which is passed through the non-linear mapping Often, a variety of model structures are evalu-
f(u) to give the input v∈ℜ of the linear dynamic ated, and the most successful one is retained. In
block. A, B, C, and D are the system matrices this section, we will first describe the estimation
(of conformal dimensions) of the linear dynamic problem in general, and then concentrate on the
block, x(k+1|k)∈ℜ is the state at time k+1 cal- pragmatic guidelines to select a model.
culated at time k, w∈ℜ is the output of the linear The general problem of parameter estimation
block which is passed through the non-linear is formalized as follows: Let the model’s general
mapping h(w) to give the output y∈ℜ of the plant. mathematical structure be represented by an op-
The static non-linear functions f(u) and h(w) are erator, M. Let the model depend on a set of pa-
assumed to be invertible. The H-W model can be rameters, ordered in a vector, θ. Then, for a
used to investigate whether it is possible to infer specific parameter vector, θ0, the y(t,θ0)=Μ(θ0,
STN spike trains using only the underlying local u(t)) is a static input/output function or a transfer
field potentials (LFPs) from intranuclear record- function in the Laplace domain, where u is the
ings, acquired intraoperatively during deep brain input and y is the output. Now, if the model struc-
stimulation procedure. The model regards the LFPs ture, M, and the parameter vector, θ0, exactly
to be the input, and the presence of the spikes to represent the physical system, the objective of
be the output of a Hammerstein-Wiener model system identification is then to find a suitable
and predicts, at least partially, that STN spikes model structure, M, and corresponding parameter
can indeed be inferred from intranuclear LFPs, vector, θ, given measurements of input and output.
at least with moderate success. Such a model can The identified model will have a parameter vec-
be seen in Figure 9. tor, 
q 0 , and generate yˆ(t ) = M (qˆ , u(t )) , where
0

yˆ(t ) is an estimate of the system output, y(t). The

system identification problem is then to choose
17.10. IDENTIFICATION OF
the model structure model, M, and find the cor-
PHYSIOLOGICAL CONTROL
SYSTEMS responding parameter vector,  q , that produces
0

the model output that best predicts the measured

The system identification approach to constructing system output.
a mathematical model of a physiological system In the remaining parts of this section, the
is much different than what has been presented identification steps that are usually involved for
until now. The modeler’s task is first to select a discrete models are presented. The process requires
general form, or structure, for the mathematical four steps, which are often applied iteratively:

Figure 9. A Hammerstein-Wiener cascade model is able to predict the spikes from the recorded local
field potentials (Michmizos & Nikita, 2010).

773
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

1. Postulate a model form (structure) and select (VAF) is high enough). If some parameters
the appropriate identification tool. have confidence levels which include zero,
2. Postulate a model order and imbed data in then an attempt should be made by the in-
a set of equations for the identification. vestigator to fit another model with those
3. Compare predictions to real observations in parameters removed (if one is manipulating
the data set used for identification (i.e., find his/her own regressor matrix); otherwise, a
residuals and their statistics), and estimate fit with a supplied algorithm setting a lower
confidence in parameter estimates; then, order is to be attempted. Once an order for
correct model form or order as needed and the current model form is decided, the final
repeat steps i-ii. parameter estimates must then come from
4. Validate the selected model by examining a final fit with that selected order.
predictions in new data sets in the same 2. Residual is not white and not zero-mean:
experiment or in completely novel experi- Assuming the underlying noise statistics are
mental protocols. If several model forms indeed Gaussian, this means that either the
perform equivalently in step ii, they may wrong model form (schematic or hypotheti-
not do so here when tested on new data. cal relationship) is attained, or an insufficient
number of parameters exist. The investigator
First, a model order is selected, by fixing the should then increase the order and try again.
model type and polynomial orders. The properties Subsequently, the investiagator is to examine
of the residual noise, r = y − yˆ , y being the real the current residuals for deterministic trends
output and ŷ being the predicted output, for the (like ramps or sinusoids), and adjust the
data set can be examined. For standard regression, model form accordingly.
if the residual is white with approximately Gauss- 3. Residual is not Gaussian but is zero-mean
ian distribution, its variance can be used to set and white: This may happen if the underly-
confidence intervals on the parameters and decide ing noise properties are actually themselves
if any of them are superfluous. If the noise sequence not Gaussian, or the wrong model order
is not white, or diverges greatly from the normal exists. Whatever the reason, one cannot rely
distribution, then it could be assumed that we on the usual t-statistics for the confidence
have either the wrong model form or the wrong intervals of estimated parameters – these
order in the current form. could lead to erroneous selection of model
Pragmatic guidelines to select model structures order and/or pertinent parameters. In this
at this stage are: case, it is often recommended to resort to
‘Monte Carlo’ or ‘Bootstrap’ methods.
1. Residual is nearly white, Gaussian and These approaches are computationally de-
zero-mean: The t-statistics should be used manding, but they generate pseudo statistics
in order to define confidence intervals on on parameter estimates from which more
all the estimated parameters, or to examine accurate confidence intervals can be deter-
those provided by the applied estimation mined, regardless of the form of each pa-
function. If all are significantly different rameter’s probability distribution (e.g., limits
from zero at the desired confidence level, for 95% of area under curve). Monte Carlo
then the current model is a valid possibil- relies on repeating the estimation routine
ity, provided the quality of fit is satisfac- many times, using many experimental pro-
tory (e.g., the %Variance of Accounted for tocols, or dividing a large data set into
multiple sets – but, it may not always be

774
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

feasible to have long experimental protocols. physiological system with a linear model
Bootstrap instead uses a single data set and y = ∑ i wi ui . Please note that the real
generates multiple sets for parameter estima- system may or may not be linear, and that
tion by for example: A) iteratively using the data we acquire from the experiments
each estimate to generate a new shuffled are usually noisy. But if we insist on finding
noise sequence (the investigator is to use the optimal linear model according the least
r = y − yˆ as defined previously, shuffle square criterion, we have to find an optimum
randomly, and add back onto ŷ for a ‘new’ matrix notation, WT for which y=WTU, U
data set), creating new virtual noisy data being the input matrix. According to the least
sets-, or B) selecting a subset of data ran- square criterion, we have to calculate
domly from the original set to generate es- W=Φ-1P, whereby P = E [Y∙U] and Φ = E
timates with each. This is repeated as many [U∙UT], and where E stands for expectation
times as necessary to obtain smooth param- and UT is the transpose input matrix. For
eter histograms; and, does not lengthen further information on the origin and proof
experimental data acquisition. However, pertaining to the above discussion, the
there are differences between approaches A reader is to refer to a textbook on linear
and B. In particular, method B means that parameter estimation, e.g., (Ljung, 1999).
each estimation run will have fewer data
observations than that of the original experi-
mental data length; in method A, the number Assume that we have a static system with two
of observations entering the estimation step inputs and one output, y=w1∙u1+ w2∙u2. To estimate
is always the same total as the whole data the model’s parameters, we conduct an experiment
segment. measuring the inputs and the output four times,
as shown in Table 1.
Finally, validation of the selected model should We now calculate the estimation of P as fol-
include a demonstration that the predictors perform lows:
well on new data not used in the original fitting
procedure. This can be data reserved from the y ⋅ u  1  5.1 − 1.1 − 0.8 + 4.7  1.975
P = E Y ⋅U  = E  1
= 
 =
 


original experiment, or a totally new data stream y ⋅ u2  4 5.1 + 1.1 + 0.8 + 4.7  2.925
from a different protocol. The best models will
fit well the data used in the identification, and
will also duplicate well other data sets. This last Next, we calculate the estimation of Φ as
cross-validation step is the final test which can tell follows:
the best models from those specific to a special
condition. Hence, this is an important step in jus-  u ⋅ u u ⋅ u  1 0
Φ = E U ⋅U T  = E  1 1 1 2  
tifying the final choice of a model form and order.   u ⋅ u u ⋅ u  = 0 1
 2 1 2 2   
17.10.1 Applied Example
Now, we can calculate the optimal parameters
Example 17.8: Next, we will present a simple
W =(Φ-1∙P)T=[1.975, 2.925].
T
parameter estimation problem for a linear
We can see that the model’s outputs for the four
model, in order to illustrate the theory previ-
experimental measurements in Table 2.
ously discussed. Let us model an unknown

775
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

Table 1. System’s inputs and outputs. Table 2. Model’s outputs.

u1 u2 y
u1 u2 ŷ
-1 -1 -5.1
-1 1 1.1 -1 -1 -4.90

1 -1 -0.8 -1 1 0.95
1 1 4.7
1 -1 -0.95

1 1 4.90

As one can see, the model outputs are very

similar to the data we acquired. However, in a
real experiment, it is highly expected to have more 17.11.1. Basic Principles
noise than the one we had here. Hence, more
experiments should be conducted to gather more The term neural network was traditionally used
data and have a good estimation of the model’s to refer to a network or circuit of biological neu-
parameters. rons. The usage of the term has changed to refer
to artificial neural networks, which are composed
of artificial neurons or nodes. Various neural
17.11. ARTIFICIAL NEURAL network algorithms currently exist, but they all
NETWORKS FOR PHYSIOLOGICAL share common characteristics that include a set of
SYSTEMS CONTROL inputs and outputs, the distributed processing of
the information, and their adaptive parameteriza-
Artificial neural network models represent a tion. The structure of a neural network resembles
black box type of model. These models are used the structure of the nervous system. The input
in situations where the precise functioning of the information inserted into such a network is col-
system is not understood or easily implemented, lectively processed by a group of distinct units
and only the sample input-output data are known. (in analogy to the neurons). Each processing unit
This section will provide a brief description of the interacts with the information given locally, and
basic principles of neural network control systems then sends an output to other units or the environ-
and their use in control of physiological systems. ment (output information). The significance of a
Neural networks have been used for more than certain connection (synapse) between two units
two decades in solving engineering problems, is determined by a value of strength (synaptic
especially in pattern recognition and pattern clas- weight). These values modify the input-output
sification applications. Neural networks are also behavior of the entire neural network, and are
used in modeling problems that are difficult to adjusted according to a learning algorithm. In order
solve. For instance, controlling a nonlinear system to design a neural network, one has to consider
has always been an advanced modeling task that the internal characteristics, the architecture, and
most of the times led to an insufficient solution. the number of the processing units, as well as the
The introduction of the neural networks to the field learning algorithm.
of physiological systems control resulted in a new The architecture of the neural network is not
area of research for both the neural network and the only analogy between the artificial and the
systems control scientific communities. biological systems. The internal characteristics of
the processing unit mimic the ones of a neuron. A
neuron receives chemical messages (inputs) from

776
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

other neurons that are transformed to dendritic where, n is the number of neurons that give their
potential, which is then added up in the neuron’s outputs to the neuron, and wi is the synaptic weight
soma to fire an action potential (output). In general, from presynaptic neuron i to the postsynaptic
the decision upon firing an action potential relies neuron. Equations 17.40 and 17.41 denote that
on a nonlinear function of a weighted summation the output of an artificial neuron depends on its
of the neuron’s inputs. The most common equation inputs only and does not depend on time; hence,
used to model the decision is the sigmoid curve the output is a static nonlinear function of the
produced by the mathematical function having an weighted summation of the inputs.
“S ” shape, as shown in Figure 10. The general The most common architecture used in neural
equation for a sigmoid function is networks is a structure that uses three layers of
processing units. The first layer, the input layer,
1 processes the input information and sends its
Y = (17.40)
(1 + e −mx ) output to a second layer, called the hidden layer
that sends the processed information to the last
layer, the output layer. A neural network is called
where, x is the input (the weighted summation of feedforward network if all its processing units
the artificial neuron), and m is a constant that receive inputs from the units of previous layers.
regulates the slope of the sigmoidal output func- Defining the number of the processing units on
tion. For m = 1, the function is named logistic each layer is more of an art than a science.
function and is related to population growth stud- A general learning algorithm, used to train a
ies. Sometimes, a constant is added to the term neural network, is a function of: i) the learning
–mx, and is called the bias of the sigmoid function. rate, η; ii) the activation of the presynaptic unit,
The inputs of an artificial neuron, ui, are related ai, and that of the postsynaptic unit, aj; and, iii) a
to their weighted summation by the equation training (error in supervised learning techniques)
signal, eij
n
x = ∑ w i ui (17.41)
i =1 ∆wij = f (h, ai , a j , eij ) (17.42)

Figure 10. Sigmoid function for various values of m

777
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

Not all learning algorithms, used in practice, efficient. The neural network is trained using
include all those parameters. For instance, a Heb- learning data acquired from the system’s output.
bian learning algorithm changes the weights, wij, Alternatively, computer simulations of the true
in proportion to the product of the presynaptic system can be used. The training (error) signal, eij,
activation, ai, and the postsynaptic activation, is usually the difference between the output of the
aj. Another class of learning algorithms that is original controller and the output of the network.
heavily used is the one that uses gradient descent After the network is adequately trained, it can
techniques to adjust the synaptic weights, wij. An replace the controller entirely. In direct inverse
example is the error back-propagation algorithm control systems, shown in Figure 11b, the neural
that uses an error gradient descent technique. This network is used to estimate an inverse model of
technique passes the output error to previous layers the system to be controlled. The network learns
of a neural network in order to estimate the input to map the output of the system to its input. Di-
signal for any given neuron. These techniques are rect inverse control systems are common among
also classified as supervised learning techniques physiological control systems.
since they use a specification of the true output in
order to estimate the output error of the network 17.11.2. Applied Examples
prediction. Other learning algorithms, such as
reinforcement learning, are used when the true Example 17.9: We will now use the basic notions
output of the network is not directly accessible. described in previous section to construct a
The motivation behind the utilization of neural basic model for associative memory, which
networks to control a system is usually our need to stands as well, as the most likely model for
model a nonlinear process or the requirement for cognitive memories. It is based on the obser-
the control system to adapt. In a control system, vation that human beings retrieve informa-
the neural network mimics the behavior of one tion best when it is linked to other related
or more of the system’s components so well that information. That “linkage” between already
it can even replace them. In supervised control known and new information is mathemati-
systems, shown in Figure 11a, a neural network cally described by the weight (strength) of
may replace the controller of the system in situa- the connections between processing units of a
tions where the true controller is not time or cost neural network. The architecture, illustrated

Figure 11. Single modules of controlling structures can be replaced from neural networks models

778
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

in Figure 12, is the most general static linear where, e is a constant called the learning rate,
neural network since the addition of neuron usually taken to be the reciprocal of the number of
layers does not change the capability of the training vectors (usually referred to as the learning
structure. examples) presented. In this case, the weights of
the neuron will be

In this example, the input can be inferred as a set 1 × 1 + (−1) × (−1) 1/2
of characteristics of an object (e.g., a set of mea- 1 1 × 1 + (−1) × (−1) 1/2
surements that describe features of a tumor in a {
W = e∑ i =1 u =
L
}
4 (−1) × 1 + 1 × (−1)
=
−1 / 2
CT image, such as tumor diameter, number of (−1) × 1 + 1 × (−1) −1 / 2
tumors found, level of seriousness with respect
to location, etc.) and the output can be inferred
as a decision (e.g., the degree of malignancy, the
Now, we question ourselves about what will
prognosis of the disease, etc.) For convenience
be the result, y, if a new, unseen before, input is
reasons, we select the inputs and the outputs of
introduced into the neural network. Let us take
the neural network to be binary {-1, 1}. Let us
for example, the input u = [1, 1, 1, −1] . For this
assume that a learning set of two inputs-output is
v e c t o r, t h e o u t p u t i s c a l c u l a t e d a s
given to train the neural network. Let the first
y = ∑ i wi ui = 1 / 2 + 1 / 2 − 1 / 2 + 1 / 2 = 1 . As
input be u 1 = [1, 1, −1, −1] and let the second
one can see, the result of what we acquire when
input be u 2 = [−1, −1, 1, 1]. The respective outputs
u 1 is the input of the neural network, which is
are y 1 = [1] and y 2 = [−1]. According to Hebb’s what we really wanted, since the new item is
rule, the weights represent the correlation between closer to the first learning example (only one bit
the input and the output needs to be inversed to have an identical input,
compared to three bits in the second learning
wi = ε∑ λ=1 x iλyiλ
L
example).

Figure 12. Associative linear neural network

779
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

However, the associative memory has many rate is given by trapezoidal or triangular function.
disadvantages. A major drawback is that the stored The used RNN is a fully connected multilayered
data should be binary orthogonal vectors. Another perceptron NN with two recurrent loops, the
one is that there are a lot of intra-neuron connec- initial weights of which are set to unity. The RNN
tions. Other nonlinear associative memories exist; is trained using the Real Time Recurrent Learning
however, they are not as simple as the example (RTRL) algorithm (Williams & Zipser, 1989),
given and they are beyond the scope of this chapter. which is a sequential, error-correction learning
based algorithm and allows the RNN to update
Example 17.10:Figure 13 presents a personal- the weights, while operating, as long as the RNN
ized insulin infusion advisory system (IIAS) is provided with the correct glucose level value.
which serves as a control algorithm towards The teacher-force version of the RTRL has been
the development of a closed-loop artificial applied, according to which the RNN replaces the
pancreas using the subcutaneous (SC) route previous glucose level prediction with the cor-
(Mougiakakou et al., 2010). The IIAS is responding glucose level value, when available,
able to provide real time estimations of the in order to perform future predictions.
appropriate insulin infusion rate for type 1 NMPC: The NMPC uses the personalized glu-
diabetes mellitus (T1DM) patients using con- cose-insulin metabolism model, which provides
tinuous glucose monitors and insulin pumps. estimates of the future glucose levels. The NMPC
It is based on Nonlinear Model Predictive is based on an optimizer, which computes at each
Control (NMPC) and comprises of two sample time future control movements based on
modules: i) a personalized glucose-insulin the minimization of an appropriate cost function.
metabolism model, based on the combined Particularly, at each sample time: i) future outputs
use of a Compartmental Model (CM) and a are generated by the personalized glucose-insulin
Recurrent Neural Network (RNN), and ii) an metabolism model; ii) a cost function of the future
NMPC strategy. For the in silico evaluation control movements is minimized, providing a set
of the IIAS, a Mathematical Model (MM) of of future control signals; and, iii) only the first ele-
a patient with T1DM has been used. Each ment of the suggested control sequence is applied
of the aforementioned modules is briefly to the system. The cost function encompasses the
described in the following. differences between the glucose predictions and
the desired glucose level.
Personalized glucose-insulin metabolism MM of a patient with T1DM: The MM of a Type
model: The model, which is based on the combined 1 diabetes patient consists of the following CMs:
use of a CM and an RNN, is able to provide glu- i) an SC insulin absorption model, ii) a glucose
cose predictions (Zarkogianni et al., 2007; Mou- metabolism model, iii) a SC glucose absorption
giakakou et al., 2008). More specifically, informa- model, and iv) a model for the glucose absorption
tion regarding meal intake is fed to the CM, which into the blood from the gut.
simulates the glucose absorption into the blood
from the gut. CM’s output along with the SC
insulin intake and previous SC glucose measure- 17.12. MODELING CHAOS
ment are applied to the RNN, which models the IN PHYSIOLOGY
patient’s glucose kinetics and predicts subsequent
glucose levels. The CM for glucose absorption Reductionists’ approach treated the body as a
into the blood from the gut is linear and consists machine in which the relationships among the
of one compartment, while the gastric emptying subsystems were governed by deterministic laws.

780
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

Figure 13. A multicompartmental model of artificial pancreas that is based on neural networks

Current research has proven that for a liveable dict accurately the electrical activity in a certain
system to maintain its milieu intérieur, the internal area of the brain, one has to have a complete and
environment, a plethora of interrelated feedback precise description of everything that would have
loops are miraculously put in place and in balance. an effect on that particular brain area. It is logical
On the other hand, the phenomenological functions to assume that for a human observer, the factors
of a biological subsystem, especially the ones contributing to the area’s activity are infinite. What
observed macroscopically, seem aperiodic and is more, each one of these factors plays a role in
unpredictable in nature. The biological signals are creating the area’s activity. That explains why an
so variable that they appear as random or noisy. activity recorded from the brain of an individual
To illustrate this “stochastic determinism”, we never looks the same, even if the subject repeatedly
consider a large cruising boat, full of passengers. executes the same function. Another characteristic
Any one of the passengers (processes) is free to of such systems is the presence of order under the
wonder around the boat (system), whereas the absence of periodicity. The output of a chaotic
boat itself has a determinate route, regardless of system, although follows a general pattern (called
the random movements of its passengers. The strange attractor), it is random and never repeats
paradigm illustrates that we are incredibly ordered itself. Another characteristic aspect is the ability of
on several levels, but irregularly so. The human these systems to fall into the chaotic behavior and
body is not a deterministic machine, but an amaz- come out of it, depending on the situation. When
ingly complex chaotic system. the system instability becomes large enough, the
Chaos (χάος) is an ancient Greek word given to system splits and returns to order (the analysis
someone to show that he was preponderant of all of such behaviors is called bifurcation analysis).
the others. Similarly, in science, chaos describes a A chaotic system, although deterministic in its
deterministic system that is extremely complicated structure, appears to be extremely variable. The
for its observer to be fully understood. From the structure of a chaotic system is not required to be
point of view of an observer with limited capabili- complex. In fact, simple nonlinear deterministic
ties on data selection and information understand- systems can exhibit chaotic behavior. For example,
ing, a chaotic system is an inherently unpredictable chaotic solutions to cellular membrane equations
system due to its extraordinary sensitivity to its have been found (Chay, 1985).
internal conditions. For instance, in order to pre-

781
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

The heart is one of the prime chaotic physi- varying its rhythm, the heart relaxes for differ-
ological systems (Biktashev & Holden, 1998; ent time periods per beat; this limits its fatigue.
Belair & Glass, 2003). A physician may judge upon Also, a chaotic system shows better adaptation
the healthy behavior of the heart by its periodic capabilities. The heart is able to compensate for
beats. However, our hearts almost never beat the varying blood demands. From a person dreaming
same way twice. A more thorough study reveals of playing a soccer game to someone actually
a varying interval between beats. More than one running in a soccer field for 90 minutes, it is the
reason exists for this variability. The natural pace- variance in beat rhythm and intensity that makes
maker of the heart, named the sinoatrial node (SA the heart effective at any of the conditions met
node) and found in the right atrium of the heart, in an unknown external environment. When the
is a group of cells that generates the normal sinus body’s demands for blood increase, the heart is
rhythm. Stimulation of the parasympathetic fibers able to pick up the slack without the shock of a
that reach the SA node causes a decrease in the beat quick tempo change (Ward, 2001). In the brain,
rate. On the other hand, stimulation of sympathetic chaos is related to the ability to learn. A never
fibers that reach the SA node causes an increase seen before stimulus in the brain, moves the
in the SA node rate, and a subsequent increase underlying subsystems to an unpatterned chaotic
of both the heart rate and the force of the heart state. This chaos results in the ignition of a new
contraction. The existence of the two antagonistic network assembly that is specifically associated
systems (sympathetic and parasympathetic) cre- to the new stimuli. A chaotic system is also able
ates the diversity observed in the temporal distance to reach new solutions. Such a system is able to
between two subsequent beats. In addition, a third learn from its mistakes and create new pathways to
system, the respiratory system, further increases deal with old problems. Thus, what was regarded
the heart rhythm variability since the beating of as randomness in the brain, started to be proved
the heart increases with increased inspiration. as an essential part of normal brain function.
However, the best place for someone to search Although the first and most general single
for chaotic behavior is the human brain. The word definition of health was “balance”, it seems
fundamental reductionist approach, proposed in that “out of balance” situations inside the body
1891 by H. Waldeyer-Hartz, regarded the brain are also connected to health. If we introduce to
functions to be fully modeled in the level of dis- a linear system an input that is slightly out of its
crete individual neurons. The neuron doctrine, typical input range, the system’s output will most
as this fundamental idea was named, is strongly probably be derailed. A nonlinear system, even if
opposed by modern chaos theory. The brain pos- it sees at its input a “bizarre” nudge, it will most
sesses a large number of feedbacks that give rise to probably return to its starting point. Let us look
internal uncertainties amplified over time, making to what is happening in a diseased body. Take for
long term predictions of brain activity impossible example Parkinson’s disease and the basal ganglia
(Skarda & Freeman, 1990). system that controls motion. The amount of chaos
A question arises on whether the chaotic in the Parkinsonian brain actually decreases as
behavior observed in physiological systems is the loss of dopamine (a neurotransmitter used
happening by accident or on purpose. It seems in synapses) forces neurons in the basal ganglia
that there are several deterministic reasons for the system to fire in synchrony. This synchrony is
existence of randomness in the biological systems. present in recordings and results in a beta band
Take for example the heart we discussed previ- peak, observed in the local field potentials. The
ously. There is more than one good implication peak is considered to emerge as the projection of
of the variations observed in heart rhythm. By widespread synchronized beta band oscillations of

782
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

the underlying neuronal elements (Boraud et al., ological system has significant potential to become
2005; Brown & Williams, 2005). From Parkinson’s feasible, and indeed some preliminary studies
disease to seizures, disease is recognized as an have shown significant improvement in body-
acute attack of order against chaos. Physicians prosthetics that are controlled by models (see for
have begun to classify a new order of “dynamical example Song et al., 2007; Lebedev & Nicolelis,
diseases” caused by abnormally periodic order. 2006). To successfully implement this combined
Epileptic seizures, Parkinson’s disease, heart at- approach, it is essential that the mathematical
tack, and infant apnoea are just a few such dynamic models are sophisticated enough to capture the
disorders. Even aging itself is related to a loss key physiological features of the system. This is
of deterministic variability (Kaplan et al., 1991; a computational challenge in its own right; our
Kim & Stringer, 1992). In fact, neurosurgeons body has anisotropic and multi-scale properties
are creating chaos in the brain as a form of treat- that must be realized in mathematical models and
ment of symptoms. Take, for example, the Deep solved on real time simulations. In addition, the
Brain Stimulation procedure used in Parkinsonian personalized physiological properties of each data
patients. A stimulation lead is inserted into the set should be reflected to a change of parameters
brain to deliver an electrical impulse and return in the mathematical models; and, accordingly, the
the brain to its previous chaotic state. It has re- physiological models must be customized through
cently been found that the stimulation of the STN inputting patient-specific structural and functional
results in the loss of beta synchronization in the information. Within initiatives such as the Physi-
neurons inside the nucleus (Bronte-Stewart et al., ome and Virtual Physiologic Human projects,
2009). This is not the only application of chaos the need to have universal simulation platforms,
in medicine. The opportunities are as infinite as software languages, and in general the necessity
the dynamic systems themselves. to speak the same “modeling language” became
apparent. It is also important to speak specifically
for the brain. For the first time in history of man-
17.13. THE FUTURE OF kind, the human brain initiated a discussion with
PHYSIOLOGICAL SYSTEMS itself. In this endeavor, it is extremely important
MODELING, SIMULATION, to mention the requirement to develop more ad-
AND CONTROL vanced statistical techniques applied specifically
to brain modeling. The long-term aim should be
Physiological modeling is increasingly providing a the embracement of the power of modeling and
sophisticated set of tools for processing measure- the integration of simulations with clinically,
ment inputs into clinically relevant outputs. Based scientifically, and economically effective data
on a physical and biological understanding of the acquiring techniques in order to achieve the goal
underlying processes, models have the short-term of personalized treatment. Physiological models
potential to be used to extract information that is that are able to combine patient specific data with
not directly available from the data itself, and, the personal opinion of a physician can become
thus, aid clinical diagnosis. However, various a pivotal point in the healthcare system in terms
challenges remain to be met in order to reach a of both prognosis and diagnosis. This will further
level of modeling that would take full control increase the opinion of the society that science
of a physiological system. Substituting a physi- comes not only from but also for the human kind.

783
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

17.14. PROFESSIONAL SOCIETIES Belair, J., & Glass, L. (2003). Introduction to

AND ORGANIZATIONS dynamics in nonlinear difference and differential
equations. In Beuter, A., Glass, L., Mackey, M.
Engineering in Medicine and Biology Society C., & Titcombe, M. S. (Eds.), Nonlinear dynamics
(EMBS) in physiology and medicine (pp. 9–34). Berlin,
www.embs.org Germany: Springer-Verlag.
NSR Physiome Project, National Simulation
Biktashev, V. N., & Holden, A. V. (1998). Re-
Resource, Department of Bioengineering,
entrant waves and their elimination in a model of
University of Washington, Seattle, WA, USA
mammalian ventricular tissue. Chaos (Woodbury,
http://www.physiome.org/
N.Y.), 8(1), 48–56. doi:10.1063/1.166307
Virtual Physiological Human Network of Excel-
lence Boraud, T., Brown, P., Goldberg, J., Graybiel,
http://www.vph-noe.eu/ A., & Magill, P. (2005). Oscillations in the basal
ganglia: The good, the bad, and the unexpected.
The Basal Ganglia, 3-24.
17.15. CHAPTER SUMMARY
Bower, J. M., & Beeman, D. (1998). The book of
GENESIS: Exploring realistic neural models with
In this chapter, a variety of techniques to model
the general neural simulation system (2nd ed.).
physiological systems and study their underlying
Berlin, Germany: Springer-Verlag.
functions are described. The potential and limita-
tions of the presented methodologies are discussed Bronte-Stewart, H., Barberini, C., Koop, M.,
and supported with appropriate examples. Com- Hill, B., Henderson, J., & Wingeier, B. (2009).
partmental analysis describes a biological system The STN beta-band profile in Parkinson’s disease
with a finite number of compartments. Almost all is stationary and shows prolonged attenuation
biological systems are inherently nonlinear, and a after deep brain stimulation. Experimental Neu-
purely linear model is, thus, partially satisfactory. rology, 215(1), 20–28. doi:10.1016/j.expneu-
However, linear models show important advan- rol.2008.09.008
tages due to their simplicity. Other approaches,
Brown, P., & Williams, D. (2005). Basal ganglia
such as nonlinear models and neural networks, may
local field potential activity: Character and func-
lack the theoretical foundation upon which linear
tional significance in the human. Clinical Neuro-
modeling of physiological systems is based, but
physiology, 116(11), 2510–2519. doi:10.1016/j.
promising theoretical developments have attested
clinph.2005.05.009
the importance of these techniques for successful
simulation and control of biological processes. Chay, T. R. (1985). Chaos in a three-variable
model of an excitable cell. Physica D. Nonlinear
Phenomena, 16(2), 233–242. doi:10.1016/0167-
REFERENCES 2789(85)90060-0

Beeler, G. W., & Reuter, H. (1977). Reconstruction Enderle, J. (2005). Physiological modeling.
of the action potential of ventricular-myocardial fi- In Enderle, J., Bronzino, J., & Blanchard, S.
bres. The Journal of Physiology, 268(1), 177–210. (Eds.), Introduction to biomedical engineering
(pp. 693–722). Burlington, MA, USA: Elsevier
Academic Press.

784
https://www.studyc.info/
Physiological Systems Modeling, Simulation, and Control

Furley, D., & Wilkie, J. (1984). Galen on respi- Longriff, J. (1989). Pre-Socratic philosophy and
ration and the arteries. Princeton, NJ: Princeton Hippocratic medicine. History of Science, 27,
University Press. 1–39.
Goodman, D. S., & Noble, R. P. (1968). Turnover Marketos, S. G. (1997). The medical school of
of plasma cholesterol in man. The Journal of Clini- Kos, Hippocratic Medicine. Human Sciences,
cal Investigation, 47(2), 231–241. doi:10.1172/ Forum Trends in Experimental and Classical
JCI105719 Medicine, 7, 313–324.
Hales, S. (1733). Statical essays: Containing Michmizos, K. P., & Nikita, K. S. (2010). Can
Haemastaticks (Vol. 2). London, UK: Innys and we infer subthalamic nucleus spike trains from
Manby. intranuclear local field potentials? Proceedings
of the 32nd Annual International Conference of
Hines, M. L., & Carnevale, N. T. (1997). The
the IEEE Engineering in Medicine and Biology
NEURON simulation environment. Neural
Society, EMBC’10, (pp. 5476-9547).
Computation, 9(6), 1179–1209. doi:10.1162/
neco.1997.9.6.1179 Mougiakakou, S., Prountzou, A., Zarkogianni, K.,
Bartsocas, C., Nikita, K., & Gerasimidi-Vazeou,
Hodgkin, A. L., & Huxley, A. F. (1952). A quan-
A. (2008). Prediction of glucose profile in children
titative description of membrane current and its
with type 1 diabetes mellitus using continuous
application to conduction and excitation in nerve.
glucose monitors and insulin pumps. Hormone
The Journal of Physiology, 117(4), 500–544.
Research, 70, 22–23.
Kaplan, D., Furman, M., Pincus, S., Ryan, S.,
Mougiakakou, S. G., Bartsocas, C. S., Bozas,
Lipsitz, L., & Goldberger, A. (1991). Aging and
E., Chaniotakis, N., Iliopoulou, D., & Kouris, I.
the complexity of cardiovascular dynamics. Bio-
(2010). SMARTDIAB: A communication and in-
physical Journal, 59(4), 945–949. doi:10.1016/
formation technology approach for the intelligent
S0006-3495(91)82309-8
monitoring, management and follow-up of type 1
Kim, J. H., & Stringer, J. (1992). Applied chaos. diabetes patients. IEEE Transactions on Informa-
New York, NY: John Wiley and Sons, Inc. tion Technology in Biomedicine, Special Issue on
New and Emerging Trends in Bioinformatics and
Krebs, H. I., & Hogan, N. (2006). Therapeutic
Bioengineering, 14(3), 622–633.
robotics: A technology push, Proceedings of the
IEEE, Special Issue on Medical Robotics, 94(9), Noble, D. (1960). Cardiac action and pacemaker
1727-1738. potentials based on the Hodgkin-Huxley equations.
Nature, 188, 495–497. doi:10.1038/188495b0
Krogh, A. (1919). The number and distribution
of capillaries in muscles with calculations of the Noble, D. (2002). Modeling the heart – From genes
oxygen pressure necessary for supplying the tis- to cells to the whole organ. Science, 295(5560),
sue. The Journal of Physiology, 52(6), 409–415. 1678–1682. doi:10.1126/science.1069881
Lebedev, M. A., & Nicolelis, M. A. (2006). Popper, K. R. (1959). The logic of scientific dis-
Brain-machine interfaces: Past, present and fu- covery. New York, NY: Basic Books.
ture. Trends in Neurosciences, 29(9), 536–546.
Schellenberger, J., Park, J. O., Conrad, T. C., &
doi:10.1016/j.tins.2006.07.004
Palsson, B. O. (2010). BiGG: A biochemical ge-
Ljung, L. (1999). System identification: Theory netic and genomic knowledgebase of large scale
for the user (2nd ed.). Upper Saddle River, NJ: metabolic reconstructions. BMC Bioinformatics,
Prentice Hall. 11, 213. doi:10.1186/1471-2105-11-213

785
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

Skarda, C. A., & Freeman, W. J. (1990). Chaos ADDITIONAL READING

and the new science of the brain. Concepts in
Neuroscience, 1(2), 275–285. Carson, E., & Cobelli, C. (Eds.). (2000). Modeling
Methodology for Physiology and Medicine. San
Song, D., Chan, R., Marmarelis, V., Hampson, Diego, CA, USA: Academic Press.
R., Deadwyler, S., & Berger, T. (2007). Nonlinear
dynamic modeling of spike train transformations Coatrieux, J. L., & Bassingthwaighte, J. (2006).
for hippocampal-cortical prostheses. IEEE Trans- Special Issue on The Physiome and Beyond.
actions on Bio-Medical Engineering, 54(6 Pt 1), Proceedings of the IEEE, 94(4), 671–677.
1053–1066. doi:10.1109/TBME.2007.891948 doi:10.1109/JPROC.2006.871765

Stewart, G. N. (1897). Researches on the circula- Deutsch, A., Brusch, L., Byrne, H., de Vries, G.,
tion time and on the influences which affect it. IV. & Herzel, H. (2007). Mathematical Modeling of
The output of the heart. The Journal of Physiology, Biological Systems: Vol. I. Cellular Biophysics,
22(3), 159–183. Regulatory Networks, Development, Biomedicine,
and Data Analysis. Boston, MA, USA: Birkhäuser.
Ward, M. (2001). Beyond chaos: The underlying
theory behind life, the universe and everything. Devasahayam, S. R. (2000). Signals and Systems
New York, NY: St. Martin’s Press. in Biomedical Engineering: Signal Processing
and Physiological Systems Modeling. Berlin,
Wiggers, C. J. (1928). The harmonic analysis of Germany: Springer.
intraventricular pressure curves. The American
Journal of Physiology, 86(2), 435–449. Enderle, J. (2010). Physiological Modeling: An
Introductory Course for Biomedical Engineers.
Williams, R. J., & Zipser, D. (1989). A learning Waltham, MA, USA: Academic Press.
algorithm for continually running fully recurrent
Neural Networks. Neural Computation, 1(2), Godfrey, K. (1983). Compartmental Models and
270–280. doi:10.1162/neco.1989.1.2.270 their Application. Waltham, MA, USA: Academic
Press.
Youm, J. B., Kim, N., Han, J., Kim, E., Joo, H.,
& Leem, C. H. (1842). … Earm, Y. E. (2006). Golnaraghi, F., & Kuo, B. C. (2009). Automatic
A mathematical model of pacemaker activity Control Systems (9th ed.). Englewood Hoboken,
recorded from mouse small intestine. Philosophi- NJ, USA: John Wiley and Sons, Inc.
cal Transactions of the Royal Society A: Math- Guyton, A. C., Jones, C. E., & Coleman, T. G.
ematical, Physical and Engineering Sciences, 364, (1973). Circulatory Physiology: Cardiac Output
1135–1154. doi:10.1098/rsta.2006.1759 and its Regulation. Philadelphia, PA, USA: W.B.
Zarkogianni, K., Mougiakakou, S., Prountzou, Saunders.
A., Vazeou, A., Bartsocas, C., & Nikita, K. S. Hall, J. E. (2011). Guyton and Hall Textbook of
(2007). An Insulin Infusion Advisory system Medical Physiology (12th ed.). Philadelphia, PA,
for type 1 diabetes patients based on non-linear USA: W.B. Saunders.
model predictive control methods. Proceedings
of the 28th Annual International Conference of Haykin, S. (1999). Neural Networks: A Com-
the IEEE Engineering in Medicine and Biology prehensive Foundation (2nd ed.). Upper Saddle
Society, EMBC’06, (pp. 5972-5975). River, NJ, USA: Prentice Hall.
Hoppensteadt, F. C., & Peskin, C. S. (2001).
Modeling and Simulation in Medicine and the
Life Sciences. Berlin, Germany: Springer.

786
https://www.studyc.info/
 Physiological Systems Modeling, Simulation, and Control

Keener, J. P., & Sneyd, J. (1998). Mathematical Pannocchia, G., Laurino, M., & Landi, A. (2010).
Physiology. New York, NY, USA: Springer- A Model Predictive Control Strategy Towards Op-
Verlag. timal Structured Treatment Interruptions in Anti-
HIV Therapy. IEEE Transactions on Bio-Medical
Khoo, M. C. K. (1999). Physiological Control
Engineering, 57(5), 1040–1051. doi:10.1109/
Systems: Analysis, Simulation, and Estimation.
TBME.2009.2039571
Piscataway, NJ, USA: Wiley-IEEE Press.
Pintelon, R., & Schoukens, J. (2001). System
(1989). InMarmarelis, V. Z. (Ed.). Advanced
Identification: A Frequency Domain Approach.
Methods of Physiological System Modeling:
Hoboken, NJ, USA: John Wiley & Sons.
Vol. 2. Published for the Biomedical Simulations
doi:10.1002/0471723134
Resource, University of Southern California. Los
Angeles, CA: USA by Plenum. Reddy, M., Yang, R. S., Andersen, M. E., &
Clewell, H. J. III. (2005). Physiologically Based
Marmarelis, V. Z. (Ed.). (1994). Advanced Meth-
Pharmacokinetic Modeling: Science and Appli-
ods of Physiological System Modeling. Berlin,
cations. Hoboken, NJ, USA: Wiley-Interscience.
Germany: Springer.
doi:10.1002/0471478768
Marmarelis, V. Z. (2004). Nonlinear Dynamic
Reddy, N. P. (2003). Modeling and simulation of
Modeling of Physiological Systems. Piscataway,
biomedical systems. In Kutz, M. (Ed.), Standard
NJ, USA: Wiley-IEEE Press.
Handbook of Biomedical Engineering and Design.
Marmarelis, V. Z., & Berger, T. W. (2005). General New York, NY, USA: McGraw Hill.
methodology for nonlinear modeling of neural
Rideout, V. C. (1991). Mathematical and Com-
systems with Poisson point-process inputs. Math-
puter modeling of Physiological Systems. Engle-
ematical Biosciences, 196(1), 1–13. doi:10.1016/j.
wood Cliffs, NJ, USA: Prentice Hall.
mbs.2005.04.002
Westwick, D. T., & Kearney, R. E. (2003).
Marmarelis, V. Z., & Zhao, X. (1997). Volterra
Identification of Nonlinear Physiological Sys-
models and three-layer perceptrons. IEEE Trans-
tems. Piscataway, NJ, USA: Wiley-IEEE Press.
actions on Neural Networks, 8(6), 1421–1433.
doi:10.1002/0471722960
doi:10.1109/72.641465
Zanos, T. P., Courellis, S. H., Berger, T. W.,
Palladino, J. (2006). Physiological Modeling,
Hampson, R. E., Deadwyler, S. A., & Marmare-
Simulation, and Control. In Bronzino, J. D. (Ed.),
lis, V. Z. (2008). Nonlinear modeling of causal
Biomedical Engineering Fundamentals. Boca
interrelationships in neuronal ensembles. IEEE
Raton, FL, USA: CRC/Taylor & Francis LLC.
Transactions on Neural Systems and Rehabilita-
tion Engineering, 16(4), 336–352. doi:10.1109/
TNSRE.2008.926716

787
https://www.studyc.info/
View publication stats
Bio-inspired computation
Biologically inspired computing (also bio-inspired computing) is a field of study that
loosely knits together subfields related to the topics of connectionism, social behavior and
emergence. It is often closely related to the field of artificial intelligence, as many of its
pursuits can be linked to machine learning. It relies heavily on the fields of biology,
computer science and mathematics. Biologically inspired computing is a major subset of
natural computation. The field of biocomputation has a twofold definition: the use of
biology or biological processes as metaphor, inspiration, or enabler in developing new
computing technologies and new areas of computer science; and conversely, the use of
information science concepts and tools to explore biology from a different theoretical
perspective. In addition to its potential applications, such as DNA computation,
nanofabrication, storage devices, sensing, and health care, biocomputation also has
implications for basic scientific research. It can provide biologists, for example, with an
IT-oriented paradigm for looking at how cells compute or process information, or help
computer scientists construct algorithms based on natural systems, such as evolutionary
and genetic algorithms. Biocomputing has the potential to be a very powerful tool.
The domain of bio-inspired computing is gradually getting prominence in the current
times. As organizations and societies are gearing towards a digital era, there has been
an explosion of data. This explosion of data is making it more and more challenging to
extract meaningful information and gather knowledge by using standard algorithms, due
to the increasing complexity of analysis. Finding the best solution increasingly becomes
very difficult to identify, if not impossible, due to the very large and dynamic scope of
solutions and complexity of computations. Often, the optimal solution for such a NP hard
problem is a point in the n-dimensional hyperspace and identifying the solution is
computationally very expensive or even not feasible in limited time. Therefore intelligent
approaches are needed to identify suitable working solutions.
In this context, intelligent meta-heuristics algorithms can learn and provide a suitable
working solution to very complex problems. Within meta-heuristics, bio-inspired
computing is gradually gaining prominence since these algorithms are intelligent, can
learn and adapt like biological organisms. These algorithms are drawing attention from
the scientific community due to the increasing complexity of the problems, increasing
range of potential solutions in multi-dimensional hyper-planes, dynamic nature of the
problems and constraints, and challenges of incomplete, probabilistic and imperfect
information for decision making. However, the fast developments in this domain is
increasingly getting difficult to track, due to different algorithms which are being
introduced very frequently. However, no study has attempted to identify these algorithms
exhaustively, explore and compare their potential scope across different problem
contexts.
In fact very few researchers are often familiar with the developments in the domain, where
more and more new algorithms are gaining acceptance and prominence. Therefore, with

https://www.studyc.info/
limited visibility across algorithms, new researchers working in this domain tend to focus
on very limited and popular approaches, and therefore often “force-fit” algorithms rather
than exploring the most suitable one, based on the problem statement, due to limited
awareness. To address this gap, we review some of the popularly used bio-inspired
algorithms as well as introduce the newly developed algorithms which have a huge
potential for applications. Further to that, we also explore the potential scope of
applications of the algorithms in specific domains, based on published scientific literature.
While twelve of the slightly popular algorithms have been discussed, the scope of future
research in other bioinspired algorithms has been discussed. However, in depth
discussion about the implementation (e.g. pseudocode, etc) and enhancements in each
algorithm is beyond the scope of the current article. Further, specific detailed citations of
each application could not be provided, but we attempt to generalize whenever possible
based on other focused reviews.

https://www.studyc.info/
 I-485/H-400 Biologically-inspired computing, Lecture Notes
Spring 2015, Indiana University, By Luis M. Rocha

1. What is Life?
“What was life? No one knew. It was undoubtedly aware of itself, so soon as it was life; but it did not know
what it was”. Thomas Mann [1924]

Threshold of Complexity
“Seeking a connecting link, they had condescended to the preposterous assumption of structureless living
matter, unorganized organisms, which darted together of themselves in the albumen solution, like crystals in
their mother-liquor; yet organic differentiation still remained at once condition and expression of all life. One
could point to no form of life that did not owe its existence to procreation by parents”. Thomas Mann [1924].

“Nothing in biology makes sense without evolution”. Theodosius Dobzhansky [1973]

Biologically-inspired computing is an interdisciplinary field that formalizes processes observed in living

systems to design computational methods for solving complex problems, or simply to endow artificial
systems with more natural traits. But to draw more than superficial inspiration from Biology we need to
understand and discuss the concept of life. It should be noted that for the most part of the history of
humanity, the question of what life is was not an important issue. Before the study of mechanics became
important, everything was thought to be alive: the stars, the skies, the rivers and mountains, etc. There was
no non-life, so the concept was of no importance. It was only when people started to see the World as
determined by the laws of mechanics that the question arose. If all matter follows simple physical laws, then
what is indeed the difference between life and non-life, between biology and physics? Let us then start with
a current dictionary definition:

“life adj.— n.1. the general condition that distinguishes organisms from inorganic objects and dead organisms,
being manifested by growth through metabolism, a means of reproduction, and internal regulation in response to
the environment. 2. the animate existence or period of animate existence of an individual. 3. a corresponding state,
existence, or principle of existence conceived of as belonging to the soul. 4. the general or universal condition of
human existence. 5. any specified period of animate existence. 6. the period of existence, activity, or effectiveness
of something inanimate, as a machine, lease, or play. 7. animation; liveliness; spirit: The party was full of life. 8.
the force that makes or keeps something alive; the vivifying or quickening principle.” [Random House Webster’s
Dictionary]

The definitions above fall into three main categories: (1) life as an organization distinct from inorganic matter
(with an associated list of properties), (2) life as a certain kind of animated behavior, and (3) life as a special,
incommensurable, quality—vitalism. Throughout this course we will see that all principles, and indeed all
controversies, associated with the study of life fall into one of these categories or the differences among them.
The third category has been discarded as a viable scientific explanation, because for science nothing is in
principle incommensurable. The question of whether life is organized according to a special design,
intelligent or mysterious, pertains to metaphysics. If the agent of design cannot be observed with physical
means, then it is by definition beyond the scope of science as it cannot be measured, and any theories derived
from such a concept cannot tested.

While metaphysical dispositions do not pertain to science, many scientists have observed that a naive
mechanistic decomposition of life may also fail to explain it. The traditional scientific approach has lead the
study of living systems into a reductionist search for answers in the nitty-gritty of the biochemistry of living
organisms. This alternative sees life as nothing more than the complicated physics of a collection of moving

https://www.studyc.info/
bodies. However, the question remains unanswered since there are many ways to obtain some complicated
dynamics, but of all of these, which ones can be classified as alive? What kind of complexity are we looking
for? No one disputes that life is some sort of complex material arrangement, but when do we reach a
necessary threshold of complexity after which matter is said to be living? Is it a discrete step, or is life a fuzzy
concept? To understand it without meaningless reduction, must we synthesize organizations with the same
threshold of complexity (first category above), or is it enough to simulate its animated behavior (second
category above)?

Information Organizes and Breeds Life

“Life is a dynamic state of matter organized by information”. Manfred Eigen [1992]

“Life is a complex system for information storage and processing”. Minoru Kanehisa [2000]

Traditionally life has been identified with material organizations which observe certain lists of properties, e.g.
metabolism, adaptability, self-maintenance (autonomy), self-repair, growth, replication, evolution, etc. Most
living organisms follow these lists, however, there are other material systems which obey only a subset of
these rules, e.g. viruses, candle flames, the Earth, certain robots, etc. This often leads to the view that life is
at best a fuzzy concept and at worst something we are, subjectively, trained to recognize—life is what we can
eat—and is thus not an objective distinction. The modern-day molecular biology view of life, on the other
hand, tends to see life as a material organization that if not completely defined by genomic information, is
at least fully controlled by it. Thus, when Craig Venter’s team [Gibson et al, 2010] recently produced a
bacteria with a “prosthetic genome” [a termed coined by Mark Bedau, see Nature | Opinion, 2010] copied
from another bacteria but synthesized in the lab, the momentous synthetic biology feat was announced as the
creation of the first synthetic or artificial life form.

The artificial life field, whose members tend to follow the fuzzy list of properties conception of life, does not
typically recognize Venter’s bacteria with a prosthetic genome as a bona fide synthesis of artificial life, since
it relies on the pre-existence of a working, naturally-obtained cell to implant a prosthetic genome into. Even
most molecular biologists will agree that we are nowhere near understanding, let alone synthesizing an
artificial cell from scratch [e.g. George Church, see Nature | Opinion, 2010]. Nonetheless, Venter’s
achievement begs at least the question of what is it about life’s design principle that makes it easier to
synthesize a working prosthetic genome than a working “prosthetic proteome or metabolome”? It also makes
us think about what does “understanding life” mean for biology, biomedical technology, artificial life, and
informatics? Why is genetic information so important and how does it relate to information technology?

Life requires the ability to both categorize and control events in its environment in order to survive. In other
words, organisms pursue (or even decide upon) different actions according to information they perceive in
an environment. Furthermore, living organisms reproduce and develop from genetic information. More
specifically, genetic information is transmitted “vertically” (inherited) in phylogeny and cell reproduction,
and expressed “horizontally” within a cell in ontogeny for the functioning of living organisms as they interact
and react with their environments—we are now sure that genetic information can also be transmitted
horizontally between organisms and play an important role in evolution [Goldenfeld & Woese 2007; Riley,
2013]. Indeed, the difference between living and non-living organizations seems to stand on the ability of
the former to use relevant information for their own functioning. It is this “relevant” which gives life an extra
attribute to simple mechanistic interactions. When an organization is able to recognize and act on aspects of
its environment which are important to its own survival, we say that the mechanisms by which the
organization recognizes and acts are functional in reference to the organization itself (self-reference). Physics
is not concerned with function. A physical or chemical description of DNA is certainly possible, but will tell

https://www.studyc.info/
us nothing as to the function of a DNA molecule as a gene containing relevant information for a particular
organism. Only in reference to an organism does a piece of DNA function as a gene (e.g. an enzyme with
some effect in an environment).

Thus it is remarkable that in Venter’s experiment, a cell with a synthesized prosthetic genome from a similar
but distinct organism, was able to reproduce over and over resulting in a cell with a different phenotype from
the original, implanted cell—in effect, a cell re-programmed by a synthesized genome. Is life then a type of
computer that can be reprogrammed? This also leads us to question how general-purpose can such genomic
re-programming be? Will it be restricted to very narrow classes of similar organisms, or will it ever be
possible to re-program any prokaryotic or eukaryotic cell ?

Emergence and Explanation

“First, nothing in biology contradicts the laws of physics and chemistry; any adequate biology must be consonant
with the ‘basic’ sciences. Second, the principles of physics and chemistry are not sufficient to explain complex
biological objects because new properties emerge as a result of organization and interaction. These properties can
only be understood by the direct study of the whole, living systems in their normal state. Third, the insufficiency of
physics and chemistry to encompass life records no mystical addition, no contradiction to the basic sciences, but
only reflects the hierarchy of natural objects and the principle of emergent properties at higher levels of
organization”. Stephen Jay Gould [1984].

This issue could be rephrased in terms of the notion of emergence. Whatever (macro-level) organization exists
after the complexity threshold for life is passed, we may say that it is emergent because its attributes cannot
be completely explained by the (micro-) physical level. In particular, function, control, and categorization
cannot be explained by the mechanics and dynamics of the components of life alone. Notice, however, that
emergence does not imply vitalism or dualism. When we say that certain characteristics of life cannot be
explained by physics alone, we mean that they must be explained by different, additional models—namely,
informational, historical and functional descriptions. In other words, though biological function, control, and
categorization cannot be explained by physics alone, organisms, like anything else, must nonetheless follow
physical laws. But information is contextual, and therefore requires more than universal models: it requires
contingent, context-specific descriptions. In particular, the origin of life, is a problem of emergence of
information from a physical milieu under specific constraints [Eigen, 1992]. This is the crux of complex
systems: the interplay between micro- and macro-level descriptions determines their behavior, and both levels
(emergence) are required to understand complexity.

The definition of emergence as an epistemological, explanatory requirement, is related to the notion of

emergence-relative-to-a-model [Rosen, 1985; Cariani, 1989] or intensional emergence [Salthe 1991]. It
refers to the impossibility of epistemological reduction of the properties of a system to its components [Clark,
1996]. As an example, we can think of phase transitions such as that of water in its transition from liquid to
gas. Water and its properties cannot be rephrased it terms of the properties of hydrogen and oxygen, it needs
a qualitatively different model. Another example of complementary models of the same material systems is
the wave-particle duality of light.

Physicists understand the laws of nature (as best they can), but it takes engineers to control nature. The very
best physicists are the very best engineers, but those are exceedingly rare (e.g. Von Neumann). The goal of
complex systems is to understand organized complexity (life, society, cognition) in the same way physicists
understand nature [Weaver, 1948]. Biology, as a discipline, has not entirely “made up its mind” if it wants
to understand life as a physicist or control it as an engineer. Due to its focus on the micro-level of life, its
biochemistry, molecular biology follows essentially a (reverse-) engineering, black-box methodology

https://www.studyc.info/
(knockouts, controls, etc.). This leads to a bit of a schizophrenic agenda: focusing exclusively on micro-level
experiments in order to suggest macro-level understandings. If the goal is control of biology, say for
biomedical advances, then we really need to focus on biotechnology engineering. If the goal is understanding,
then we need to focus more on macro-level organized complexity. Ideally, a healthy life sciences program
would tie the need to understand with the need to control better—like physicists and engineers do.

This is where complex systems, artificial life, and bio-inspired computing can contribute to a wider arena of
the life sciences; they can be used as laboratories for experimenting with theories of organized complexity,
and thus enrich our understanding of life. Artificial life concerns both the simulation and realization of life
in some artificial environment, usually the computer. At least regarding the second of its goals, artificial life
aims to understand the fundamental micro/macro-level interaction that leads to organized complexity. Bio-
inspired computing, as a more pragmatic endeavor, does not need to concern itself with synthesizing actual
life, but only with drawing analogies from life (real and artificial). Nonetheless, if the main motivation of bio-
inspired computing is that life with its designs has already solved versions of many complex engineering
problems we are interested in, then a thorough and accurate understanding of the essential characteristics of
life is inescapable. Moreover, by abstracting context-specific principles of life to make them relevant in other
settings, provides a useful laboratory to experiment with theoretical biology.

Further Readings and References:

Cariani, Peter [1989].On the Design of Devices with Emergent Semantic Functions. PhD.Dissertation. SUNY
Binghamton.
Clark, Andy [1996]. “Happy couplings: emergence and explanatory interlock.” In: The Philosophy of Artificial Life. M.
Boden (ed.). Oxford University Press, pp. 262-281.
Dobzhansky, T. [1973]. “Nothing in Biology Makes Sense Except in the Light of Evolution”. The American Biology
Teacher, March 1973 (35:125-129)
Eigen, M. [1992]. Steps Towards Life. Oxford University Press.
D. G. Gibson et al [2010]. “Creation of a Bacterial Cell Controlled by a Chemically Synthesized Genome”. Science. 329
(5987): 52-56
Goldenfeld, Nigel, and Carl Woese [2007]. “Biology’s Next Revolution.” Nature 445 (7126): 369.
Gould, Stephen Jay [1984]. Natural History; Jan84, Vol. 93 Issue 1, p24.
Mann, T. [1924]. The Magic Mountain. As quoted by Eigen [1990].
Nature | Opinion [2010] “Life after the synthetic cell”. Nature 465: 422–424
Pattee, Howard H. [1978]."The complementarity principle in biological and social structures." In: Journal of Social and
Biological Structures Vol. 1, pp. 191-200.
Polanyi, M. [1968]. “Life’s irreducible structure”. Science, 160 (3834), 1308-1312.
Riley, D. R., K.B. Sieber, K. M. Robinson, J. R. White, A. Ganesan, S. Nourbakhsh, and J. C. Dunning Hotopp [2013].
“Bacteria-Human Somatic Cell Lateral Gene Transfer Is Enriched in Cancer Samples.” PLoS Computational Biology
9 (6): e1003107.
Salthe, Stanley N. [1991], “Varieties of Emergence”. World Futures Vol. 32, pp.69-83
Schrödinger, Erwin [1944]. What is Life?. Cambridge University Press.
Weaver, W. [1948]. "Science and Complexity". American Scientist, 36(4): 536-44.

For next lectures read:

Dennet, D.C. [2005]. "Show me the Science". New York Times, August 28, 2005.
Gleick, J. [2011]. The Information: A History, a Theory, a Flood. Random House. Chapter 8.
Kanehisa, M. [200]. Post-genome Informatics. Oxford University Press. Chapter 1, Blueprint of life, pp. 1-23.
Langton, C. [1989], “Artificial Life” In Artificial Life. C. Langton (Ed.). Addison-Wesley. pp. 1-47.
Nunes de Castro, Leandro [2006]. Fundamentals of Natural Computing: Basic Concepts, Algorithms, and Applications.
Chapman & Hall. Chapter 1, pp. 1-23.
Polt, R. [2012]. "Anything but Human". New York Times, August 5, 2012

https://www.studyc.info/
2. The logical Mechanisms of Life
“The designs found in nature are nothing short of brilliant, but the process of design that generates them is utterly
lacking in intelligence of its own”. Daniel Dennett, NY Times 2005

Life-As-It-Could-Be: but, what is non-life-as-it-could-be?

“Artificial Life [AL] is the study of man-made systems that exhibit behaviors characteristic of natural living systems.
It complements the traditional biological sciences concerned with the analysis of living organisms by attempting
to synthesize life-like behaviors within computers and other artificial media. By extending the empirical foundation
upon which biology is based beyond the carbon-chain life that has evolved on Earth, Artificial Life can contribute
to theoretical biology by locating life-as-we-know-it within the larger picture of life-as-it-could-be. [...] [AL] views
life as a property of the organization of matter, rather than a property of the matter which is so organized. Whereas
biology has largely concerned itself with the material basis of life, Artificial Life is concerned with the formal basis
of life. [... It] starts at the bottom, viewing an organism as a large population of simple machines, and works upwards
synthetically from there — constructing large aggregates of simple, rule-governed objects which interact with one
another nonlinearly in the support of life-like, global dynamics. The ‘key’ concept in AL is emergent behavior.”
[Langton, 1989, pp 1-2]
“Artificial Life is concerned with tuning the behaviors of such low-level machines that the behavior that emerges
at the global level is essentially the same as some behavior exhibited by a natural living system. [...] Artificial Life
is concerned with generating lifelike behavior.” [Langton, 1989, pp 4 and 5]

The previous quotes indicate the goals of Artificial Life according to Chris Langton: the search for complex,
artificial, systems which instantiate some kind of lifelike organization. The field is interested in both
synthesizing an actual artificial living organization, as well as simulating lifelike behavior. The first goal is
more ambitious and related to the first definition of life introduced in lecture one, while the second goal is
related to the second definition. The methodology to reach either of these goals is also in line with the notion
of emergence mentioned in lecture one: from the non-linear interaction of simple, mechanistic, components,
we wish to observe the emergence of complicated, life-like, unpredictable, behavior. Natural living organisms
are likewise composed of non-living components. As pointed out in lecture one, the origin problem in biology
is precisely the emergence of life from non-living components. The material components follow, and are
completely described, by physical laws, however, a mechanistic explanation of the overall living system is
incomplete. Similarly, in Artificial Life, we have formal components obeying a particular set of axioms, and
from their interaction, global behavior emerges which is not completely explained by the local formal rules.
Clearly, the formal rules play the role of an artificial matter and the global behavior, if recognized as life-like,
plays the role of an artificial biology.

“Of course, the principle assumption made in Artificial Life is that the ‘logical form’ of an organism can be
separated from its material basis of construction, and that ‘aliveness’ will be found to be a property of the former,
not of the latter.” [Langton, 1989, page 11]

The idea is that if we are able to find the basic design principles of living organization, then the material
substrate used to realize life is irrelevant. By investigating these basic principles we start studying not only
biological, carbon-based, life — life-as-we-know-it — but really the universal rules of life, or life-as-it-could-
be. Moreover, from a better understanding of the design principles of life, we can use them to solve
engineering problems similar to those that living organisms face [Segel and Cohen, 2001; DeCastro and Von

https://www.studyc.info/
Zuben, 2005]. Several problems have been raised regarding this separation of matter from form, or the search
for a universality without matter [Cariani, 1992; Moreno et al, 1994], which will not be discussed here. What
needs to be made more explicit is the relationship between the two distinct goals of AL.

Looking at emergent behavior, obtained from formal complex systems, in search of interesting behavior leads
to a certain circularity. If AL is concerned with finding life-like behavior in artificial, universal, systems, we
are ultimately binding life-as-could-be to the behavior of life-as-we-know-it by virtue of some subjective
resemblance. This can hardly be accepted as the search for universal principles.

“They say, ‘Look, isn’t this reminiscent of a biological or a physical phenomenon!’ They jump in right away as if
it’s a decent model for the phenomenon, and usually of course it’s just got some accidental features that make it look
like something.” [Jack Cowan as quoted in Scientific American, June 1995 issue, “From Complexity to Perplexity”,
by J. Horgan, page 104]
“Artificial Life — and the entire field of complexity—seems to be based on a seductive syllogism: There are simple
sets of mathematical rules that when followed by a computer give rise to extremely complicated patterns. The world
also contains many extremely complicated patterns. Conclusion: Simple rules underlie many extremely complicated
phenomena in the world. With the help of powerful computers, scientists can root those rules out.” [J. Horgan,
Scientific American, June 1995 issue, “From Complexity to Perplexity”, page 107]
“Artificial Life is basically a fact-free science”. [John Maynard Smith as quoted in Scientific American, June 1995
issue, “From Complexity to Perplexity”, by J. Horgan, page 107]

The problem is that Artificial Life must be compared to something, otherwise it becomes a factless
manipulation of computer rules with subjective resemblances to real life. Again, we are faced with many
possible types of emergent complex behaviors, this time formal, but what kinds of behaviors can be classified
as “life-as-could-be”? What is the formal threshold of complexity needed? In the natural world we are able
to distinguish life from non-life, biology from physics due to the known signatures of bio-chemistry. In the
logical realm, we likewise need a formal criteria to distinguish logical life from logical non-life, artificial life
from artificial physics.

“Artificial Life must be compared with a real or an artificial nonliving world. Life in an artificial world requires
exploring what we mean by an alternative physical or mathematical reality.” [Pattee, 1995]

The two goals of AL are usually described as hard and soft AL respectively. The first concerns the synthesis
of artificial life from computational or material (e.g. embodied robotics) components. The second is interested
in producing life-like behavior and is essentially metaphorical. To be accepted as a scientific field, Alife
cannot settle for subjective rules of what constitutes living behavior. Indeed, whether we want to synthesize
life or merely simulate a particular behavior of living organisms, we need investigate the rules that allow us
to distinguish life from non-life . Only by establishing an artificial physics, from which an artificial biology
can emerge, and a theory, or set of rules, distinguishing the two, can we aim at a proper science based on fact.
In other words, the methodology of Artificial Life requires existing theories of life to be compared against;
it can also contribute to the meta-methodology of Biology by allowing us to test and improve its theories
beyond the unavoidable material constraints, such as the incomplete fossil record or measurement of cellular
activity. Naturally, the requirements for hard AL are much stricter, as we are not merely interested in
behaviors that can be compared to real biological systems with looser or stricter rules, but the actual
realization of an artificial organization that must be agreed to be living against some theory. Soft AL, may
restrict itself to particular behavioral traits which need only to be simulated to a satisfactory degree.

https://www.studyc.info/
Simulations, Realizations, Systemhood, Thinghood, and Theories of Life
“Boids are not birds; they are not even remotely like birds; they have no cohesive physical structure, but rather exist
as information structures — processes — within a computer. But — and this is the critical ‘but’— at the level of
behaviors, flocking Boids and flocking birds are two instances of the same phenomenon: flocking. [...] The
‘artificial’ in Artificial Life refers to the component parts, not the emergent processes. If the component parts are
implemented correctly, the processes they support are genuine — every bit as genuine as the natural processes they
imitate. [...] Artificial Life will therefore be genuine life —it will simply be made of different stuff than the life that
has evolved on Earth.” [Langton, 1989, pp. 32-33]

“Simulations and realizations belong to different categories of modeling. Simulations are metaphorical models that
symbolically ‘stand for’ something else. Realizations are literal, material models that implement functions.
Therefore, accuracy in a simulation need have no relation to quality of function in a realization. Secondly, the
criteria for good simulations and realizations of a system depend on our theory of the system. The criteria for good
theories depend on more than mimicry, e.g., Turing Tests.” [Pattee, 1989, page 63]

As Pattee points out, the bottom line is that a simulation, no matter how good it is, is not a realization.
Nonetheless, it may still be possible to obtain artificial living organisms (realizations) if, from an artificial
environment, we are able to generate, in a bottom-up manner, organizations which conform to some theory
of life we wish to test. Howard Pattee [1989] has proposed that if emergent artificial organisms are able to
perform measurements, or in other words, categorize their (artificial) environment, then they may be
considered realizations. Some claim that computational environments do not allow for this creative form of
emergence [see Cariani, 1992; Moreno, et all, 1994]. In any case, whatever artificial environment we may
use, computational or material, we need a theory allowing us to distinguish life from non-life.

Related to this issue, and in the context of complex systems science, is the search of those properties of the
world which can be abstracted from their specific material substrate: systemhood from thinghood. Systems
science is concerned with the study of systemhood properties, but there may be systems from which
systemhood cannot be completely abstracted from thinghood. Life is sometimes proposed as one of those
systems [see Rosen, 1986, 1991; Moreno et al, 1994; Pattee, 1995]. The difficulty for systems science, or
complexity theory, lies precisely in the choice of the appropriate level of abstraction. If we abstract enough,
most things will look alike, leading to a theory of factless, reminiscent analogies, exposed by Cowan and
Maynard-Smith above. If, on the other hand, we abstract too little, all fields of inquiry tend to fall into
increasingly specific niches, accumulating much data and knowledge about (context-specific) components
without much understanding of, or ability to control, the (general) macro-level organization. In the context
of life, we do not want to be tied uniquely to carbon-based life, or life-as-we-know-it, but we also do not want
life-as-could-be to be anything at all. The challenge lies precisely on finding the right amounts of systemhood
and thinghood, as well as the interactions between the two, necessary for a good theory of life, real or
artificial.

Further Readings and References

Cariani, P. [1992], “Emergence and Artificial Life” In Artificial Life II. C. Langton (Ed.). Addison-Wesley. pp. 775-797.
de Castro, L. N. & Von Zuben, F. J. (eds.) [2005]. Recent Developments in Biologically Inspired Computing. Idea
Group Publishing.
Dennet, D.C. [2005]. "Show me the Science". New York Times, August 28, 2005
Langton, C. [1989], “Artificial Life” In Artificial Life. C. Langton (Ed.). Addison-Wesley. pp. 1-47.
Klir, G. [1991], Facets of Systems Science. Plenum Press. (On Constructivism pp. 12-13)
Moreno, A., A. Etxeberria, and J. Umerez [1994], “Universality Without Matter?”. In Artificial Life IV, R. Brooks and
P. Maes (Eds). MIT Press. pp 406-410

https://www.studyc.info/
Pattee, H. [1989], “Simulations, Realizations, and Theories of Life”. In Artificial Life. C. Langton (Ed.). Addison-
Wesley. pp. 63-77.
Pattee, H. [1995], “Artificial Life needs a real Epistemology”. In Advances in Artificial Life. F. Moran, A Moreno, J.J.
Merelo, P. Chacon (Eds.). Springer-Verlag.
Rosen, R. [1986], “Some Comments on Systems and System Theory”. In Int. Journal of General Systems. Vol. 13, No.1.
Rosen, R. [1991]. Life Itself: A Comprehensive Inquiry into the Nature, Origin, and Fabrication of Life. Columbia
University Press.
Segel, L.A. and I.C. Cohen [2001]. Design Principles for the Immune System and Other Distributed Autonomous
Systems. Santa Fe Institute Studies in the Sciences of Complexity. Oxford University Press.

For next lecture read:

Nunes de Castro, Leandro [2006]. Fundamentals of Natural Computing: Basic Concepts, Algorithms, and Applications.
Chapman & Hall. Chapter 7, sections 7.1, 7.2 and 7.4.
Optional: Part I of Flake’s [1998], The Computational Beauty of Life. MIT Press.

https://www.studyc.info/
3. Formalizing and Modeling the World

“When you can measure what you are speaking of and express it in numbers you know that on which
you are discoursing. But if you cannot measure it and express it in numbers. your knowledge is of a
very meagre and unsatisfactory kind.”. (Lord Kelvin)

1
The Nature of Information and Information Processes in Nature
The word information derives from the Latin informare (in + formare), meaning to give form, shape, or
character to. Etymologically, it is therefore understood to be the formative principle of something, or to
imbue with some specific character or quality. However, for hundreds of years, the word information is used
to signify knowledge and aspects of cognition such as meaning, instruction, communication, representation,
signs, symbols, etc. This can be clearly appreciated in the Oxford English Dictionary, which defines
information as “the action of informing; formation or molding of the mind or character, training, instruction,
teaching; communication of instructive knowledge”.

When we look at the world and study reality, we see order and structure everywhere. There is nothing that
escapes description or explanation, even in the natural sciences where phenomena appear sometimes
catastrophic, chaotic and stochastic. A good example of order and information are our roads. Information
can be delivered by signs. Drivers know that signs are not distant things, but they are about distant things
in the road. What signs deliver are not things but a sense or knowledge of things – a message. For
information to work that way, there have to be signs. These are special objects whose function is to be about
other objects. The function of signs is reference rather than presence. Thus a system of signs is crucial for
information to exist and be useful in a world, particularly for the world of drivers!

The central structure of information is therefore a relation among signs, objects or things, and agents capable
of understanding (or decoding) the signs. An AGENT is informed by a SIGN about some THING. There
are many names for the three parts of this relation. The AGENT can be thought of as the recipient of
information, the listener, reader, interpretant, spectator, investigator, computer, cell, etc. The SIGN has been
called the signal, symbol, vehicle, or messenger. And the about-some-THING is the message, the meaning,
the content, the news, the intelligence, or the information.

The SIGN-THING-AGENT relation is often understood as a sign-system, and the discipline that studies sign
systems is known as Semiotics. In addition to the triad of a sign-system, a complete understanding of
information requires the definition of the relevant context: an AGENT is informed by a SIGN about some
THING in a certain CONTEXT. Indeed, (Peircean) semiotics emphasizes the pragmatics of sign-systems,
in addition to the more well-known dimensions of syntax and semantics. Therefore, a complete (semiotic)
understanding of information studies these three dimensions of sign-systems:

1. Semantics: the content or meaning of the SIGN of a THING for an AGENT; it studies all aspects
of the relation between signs and objects for an agent, in other words, the study of meaning.
2. Syntax: the characteristics of signs and symbols devoid of meaning; it studies all aspects of the
relation among signs such as their rules of operation, production, storage, and manipulation.
3. Pragmatics: the context of signs and repercussions of sign-systems in an environment; it studies
1
This subsection is an excerpt of [Rocha and Schnell, 2005]

https://www.studyc.info/
 how context influences the interpretation of signs and how well a signs-system represents some
aspect of the environment.

Signs carry information content to be delivered to agents. However, it is also useful to understand that some
signs are more easily used as referents than others. In the beginning of the 20th century, Charles Sanders
Peirce defined a typology of signs:

1. Icons are direct representations of objects. They are similar to the thing they represent. Examples
are pictorial road signs, scale models, and of course the icons on your computer. A footprint on the
sand is an icon of a foot.
2. Indices are indirect representations of objects, but necessarily related. Smoke is an index of fire,
the bell is an index of the tolling stroke, and a footprint is an index of a person.
3. Symbols are arbitrary representations of objects, which require exclusively a social convention
to be understood. A road sign with a red circle and a white background denotes something which is
illegal because we have agreed on its arbitrary meaning. To emphasize the conventional aspect of the
semantics of symbols, consider the example of variations in road signs: in the US yellow diamond
signs denote cautionary warnings, whereas in Europe a red triangle over a white background is used
for the same purpose. We can see that convention establishes a code, agreed by a group of agents,
for understanding (decoding) the information contained in symbols. For instance, smoke is an index
of fire, but if we agree on an appropriate code (e.g. Morse code) we can use smoke signals to
communicate symbolically.

Clearly, signs may have iconic, symbolic and indexical elements. Our alphabet is completely symbolic, as
the sound assigned to each letter is purely conventional. But other writing systems such as Egyptian or Mayan
hieroglyphs, and some Chinese characters use a combination of phonetic symbols with icons and indices. Our
road signs are also a good example of signs with symbolic (numbers, letters and conventional shapes), iconic
(representations of people and animals) and indexical (crossing out bars) elements.

Finally, it is important to note that due to the arbitrary nature of convention, symbols can be manipulated
without reference to content (syntactically). This feature of symbols is what enables computers to operate.
As an example of symbol manipulation without recourse to content, let us re-arrange the letters of a word,
say “deal”: dale, adel, dela, lead, adle, etc. We can produce all possible permutations (4! = 4×3×2×1 =
24) of the word whether they have meaning or not. After manipulation, we can choose which ones have
meaning (in some language), but that process is now a semantic one, whereas symbol manipulation is purely
syntactic. All signs rely on a certain amount of convention, as all signs have a pragmatic (social) dimension,
but symbols are the only signs which require exclusively a social convention, or code, to be understood.

We are used to think of information as pertaining purely to the human realm. In particular, the use of
symbolic information, as in our writing system, is thought of as technology used exclusively by humans.
Symbols, we have learned, rely on a code, or convention, between symbols and meanings. Such a
conventional relation usually specifies rules created by a human community. But it can have a more general
definition:

“A code can be defined as a set of rules that establish a correspondence between two independent
worlds”. The Morse code, for example, connects certain combinations of dots and dashes with the letters
of the alphabet. The Highway Code is a liaison between illustrated signals and driving behaviours. A
language makes words stand for real objects of the physical World.” [Barbieri, 2003, page 94]

10

https://www.studyc.info/
We can thus think of a code as a process that implements correspondence rules between two independent
worlds (or classes of objects), by ascribing meaning to arbitrary symbols. Therefore, meaning is not a
characteristic of the individual symbols but a convention of the collection of producers and recipients of the
encoded information.

Interestingly, we can see such processes in Nature, where the producers and recipients are not human. The
prime example is the genetic code, which establishes a correspondence between DNA (the symbolic genes
which store information) and proteins, the stuff life on Earth is built of. With very small variations, the
genetic code is the same for all life forms. In this sense, we can think of the genetic system and cellular
reproduction as a symbolic code whose convention is “accepted” by the collection of all life forms.

Other codes exist in Nature, such as signal transduction from the surface of cells to the genetic system, neural
information processing, antigen recognition by antibodies in the immune system, etc. We can also think of
animal communication mechanisms, such as the ant pheromone trails, bird signals, etc. Unlike the genetic
system, however, most information processes in nature are of an analog rather than digital nature. Throughout
this course we will discuss several of these natural codes.

Formalizing Knowledge: Uncovering the Design Principles of Nature2

Once we create symbols, we can also hypothesize relationships among the symbols which we can later check
for consistency with what we really observe in the World. By creating relationships among the symbols of
things we observe in the World, we are in effect formalizing our knowledge of the World. By formalizing
we mean the creation of rules, such as verbal arguments and mathematical equations, which define how our
symbols relate to one another. In a formalism, the rules that manipulate symbols are independent of their
meaning in the sense that they can be calculated mechanically without worrying what symbols stand for.

It is interesting to note that the ability to abstract characteristics of the world from the world itself took
thousands of years to be fully established. Even the concept of number, at first was not dissociated from the
items being counted. Indeed, several languages (e.g. Japanese) retain vestiges of this process, as different
objects are counted with different variations of names for numbers. Physics was the first science to construct
precise formal rules of the things in the world. Aristotle (484-322 BC) was the first to relate symbols more
explicitly to the external world and to successively clarify the nature of the symbol-world (symbol-matter)
relation. “In his Physics he proposed that the two main factors which determine an object's speed are its
weight and the density of the medium through which it travels. More importantly, he recognized that there
could be mathematical rules which could describe the relation between an object's weight, the medium's
density and the consequent rate of fall.” [Cariani, 1989, page 52] The rules he proposed to describe this
relations were:

1. For freely falling or freely rising bodies: speed is proportional to the density of the medium.
2. In forced motion: speed is directly proportional to the force applied and inversely proportional
to the mass of the body moved

This was the first time that the relationships between observable quantities were hypothesized and used in
calculations. Such a formalization of rules as a hypothesis to be tested is what a model is all about.
Knowledge is built upon models such as this that sustain our observations of the World.

2
This subsection is an excerpt of [Rocha and Schnell, 2005b]

11

https://www.studyc.info/
 “While these quantities were expressed in terms of numbers, they were still generally regarded as inherent
properties of the objects themselves. It was not until Galileo took the interrelationships of the signs
themselves as the objects of study that we even see the beginnings of what was to be progressive
dissociation of the symbols from the objects represented. Galileo's insight was that the symbols
themselves and their interrelations could be studied mathematically quite apart from the relations in the
objects that they represented. This process of abstraction was further extended by Newton, who saw that
symbols arising from observation […] are distinct from those involved in representing the physical laws
which govern the subsequent motion”. [Cariani, 1989, page 52]

Figure 1: The Modeling Relation between knowledge and reality according to Hertz
(adapted from Cariani, 1989)

“In 1894 Heinrich Hertz published his Principles of Mechanics which attempted […] to purge mechanics of
metaphysical, mystical, undefined, unmeasured entities such as force and to base the theory explicitly on
measurable quantities. Hertz wanted to be as clear, rigorous, and concise as possible, so that implicit, and
perhaps unnecessary, concepts could be eliminated from physical theories, [which he thought should be based
solely on measurable quantities].” [Cariani, 1989, page 54]. Since the results of measurements are symbols,
physical theory should be about building relationships among observationally-derived symbols, that is, it
should be about building formal models, which Hertz called "images”:

“The most direct and in a sense the most important, problem which our conscious knowledge of
nature should enable us to solve is the anticipation of future events, so that we may arrange our
present affairs in accordance with such anticipation. As a basis for the solution of this problem we
always make use of our knowledge of events which have already occurred, obtained by chance
observation or by prearranged experiment. In endeavoring thus to draw inferences as to the future
from the past, we always adopt the following process. We form for ourselves images or symbols of
external objects; and the form which we give them is such that the necessary consequents of the
images in thought are always the images of the necessary consequents in nature of the things pictured.
In order that this requirement may be satisfied, there must be a certain conformity between nature and
our thought. Experience teaches us that the requirement can be satisfied, and hence that such a
conformity does in fact exist. When from our accumulated experiences we have succeeded in
deducing images of the desired nature, we can then in a short time develop by means of them, as by

12

https://www.studyc.info/
 means of models, the consequences in the external world which only arise in a comparatively long
time, or as a result of our own interposition. We are thus enabled to be in advance of the facts, and to
decide as to present affairs in accordance with the insight so obtained. The images which we here
speak are of our conceptions of things. With the things themselves they are in conformity in one
important respect, namely, in satisfying the above mentioned requirement. For our purpose it is not
necessary that they should be in conformity with the things in any other respect whatever. As a matter
of fact, we do not know, nor do we have any means of knowing, whether our conceptions of things
are in conformity with them in any other than the one fundamental respect. [Hertz, 1894 pp. 1-2]”

A model is any complete and consistent set of verbal arguments, mathematical equations or computational
rules which is thought to correspond to some other entity, its prototype. The prototype can be a physical,
biological, social psychological or other conceptual entity.

The etymological roots of the word model lead us to the Latin word “modulus”, which refers to the act of
molding, and the Latin word “modus” (a measure) which implies a change of scale in the representation of
an entity. The idea of a change of scale, can be interpreted in different ways. As the prototype of a physical,
social or natural object, a model represents a change on the scale of abstraction: certain particularities have
been removed and simplifications are made to derive a model.

In the natural sciences, models are used as tools for dealing with reality. They are caricatures of the real
system specifically build to answer questions about it. By capturing a small number of key elements and
leaving out numerous details, models help us to gain a better understanding of reality and the design
principles it entails.

Computational Models3
“Insofar as the propositions of mathematics are certain they do not refer to reality; and insofar as
they refer to reality, they are not certain”. Albert Einstein

Computation is the ultimate abstraction of a formal mathematical system, or an axiomatic system. It is defined
by the purely syntactic process of mapping symbols to symbols. Such mapping is the basis of the concept of
mathematical function, and it is all that computers do. This abstraction requires that all the procedures to
manipulate symbols are defined by unambiguous rules that do not depend on physical implementation, space,
time, energy considerations or semantic interpretations given to symbols by observers. Formal computation
is, by definition, implementation-independent.

Modeling, however, is not entirely a formal process. The Hertzian modeling paradigm clearly relates formal,
computational models to measurements of reality against which they must be validated. The measuring
process transforms a physical interaction into a symbol – via a measuring device. The measuring process
cannot be formalized as it ultimately depends on interacting with a specific (not implementation-independent)
portion of reality. We can simulate a measurement process, but for that simulation to be a model we will need
in turn to relate it to reality via another measurement. This important aspect of modeling is often forgotten
in Artificial Life, when the results of simulations are interpreted without access to real world measurements.

3
This section is indebted to many writings of Howard Pattee, including lecture notes and personal communications.

13

https://www.studyc.info/
Likewise, a computer is a physical device that implements a particular abstract computational model as
precisely as possible. Modern day computers are so successful because they can implement general-purpose
computations almost independently of their specific physics. We do not have to worry about the specific
physical architecture of the device as we compute, even though small errors in our computations do occur due
to the physical elements of the computing device.

In summary, a computation is a process of rewriting symbol strings in a formal system according to a program
of rules. The following characteristics are important: (1) Operations and states are syntactic. (2) Symbols
follow syntactical rules. (3) Rate of computation is irrelevant. (4)Program determines result, not speed of
machine (Physical implementation is irrelevant).

References
Aris, R. (1978). Mathematical modelling techniques. London: Pitman. (Reprinted by Dover).
Barbieri, M. [2003]. The Organic Codes: An Introduction to Semantic Biology. Cambridge University
Press.
Bossel, J. (1994). Modeling and simulation. Wellesley, MA: A K Peters.
Cariani, P. (1989). On the Design of Devices with emergent Semantic Functions. PhD Dissertation,
SUNY Bionghamton.
Hertz, H. [1894]. Principles of Mechanics. tr. D. E. Jones and J. T. Walley. New York: Dover 1956
Lin, C. C. and Segel, L. A. (1974). Mathematics applied to deterministic problems in the natural
sciences. New York: Macmillan Publishing. (Reprinted by SIAM).
Polya, G. (1957). How to solve it: A new aspect of mathematical method. Pricenton, NJ: Princeton
University Press.
Rocha, L.M. and S. Schnell (2005). “The Nature of Information”. Lecture notes for I101 – Introduction to
Informatics. School of Informatics, Indiana University.
Rocha, L.M. and S. Schnell (2005b). “Modeling the World”. Lecture notes for I101 – Introduction to
Informatics. School of Informatics, Indiana University.
Saari, D. G. (2001). Chaotic elections! A mathematician looks at voting. American Mathematical Society.
Also you can read: “Making sense out of consensus” by Dana MacKenzie, SIAM News, Volume
33, Number 8, October 2000.

14

https://www.studyc.info/
4. Self-Organization and Emergent Complex Behavior
Self-organization is usually understood as the process by which systems of many components tend to reach
a particular state, a set of cycling states, or a small volume of their state space (attractor basins), with no
external interference. This attractor behavior is often recognized at a different level of observation as the
spontaneous formation of well-organized structures, patterns, or behaviors, from random initial conditions
(emergent behavior). The systems used to study this behavior are referred to as dynamical systems or state-
determined systems, since every trajectory is perfectly determined by its initial state. Dynamical systems are
traditionally studied by continuous variables and sets of discrete-time difference equations (such as the
logistic map) or continuous-time differential equations (such as models of the motion of bodies under
gravitational forces). However, self-organization is more easily studied computationally with discrete
dynamical systems (DDS) such as Boolean networks or cellular automata.

The state-determined transition rules of DDS are interpreted as the laws of some physical system [Langton,
1986] where the state of each component depends on the states of its neighbor (or related) components at the
previous time instance. DDS possess a large number of components or variables, and thus very large state
spaces. However, when started with random initial conditions (note: not from special initial conditions) they
tend to converge, or self-organize, into small sets of attractor states in this space. Attractors may be chaotic
in which case the emergent behavior is sensitive to initial conditions. But even chaotic attractors tend to be
restricted to small volumes of their state-space (e.g. chaotic in a subset of dimensions of the state-space),
therefore we still consider the convergence of a dynamical system into a chaotic basin of attraction to be a
form of self-organization.

Since material systems are accurately modeled by dynamical systems, it follows from the observed attractor
behavior [Wuensche and Lesser, 1992] of these systems that there is a propensity for matter to self-organize
(e.g., [Kauffmann, 1992]). In this sense, matter is described by the (micro-level) dynamics of state transitions
and the observed (emergent or macro-level) attractor behavior of self-organization. In general, attractors
manifest or emerge as global patterns that involve many of components of the dynamical system, and are not
easily describable in terms of their state-determined transition rules. For instance, the simple transition rules
of the automata in Conway's Game of Life cannot describe what the emergent patterns of "blinkers" and
"gliders" are. These emergent patterns pertain to a different, complementary level of observation of the same
system [Pattee, 1978]. The process of self-organization is often interpreted as the evolution of order from
random initial conditions. However, notice that this evolution is limited to the specific attractor landscape of
a given dynamical system. Unless its parameters are changed (structural perturbation), no dynamical system
can escape its own attractor landscape. This limitation will become more apparent when we approach the
problem of self-replication.

Life on the Edge of Chaos?

Another interesting aspect of the behavior of dynamical systems concerns the concept of bifurcation or phase
transition. When the parameters of a dynamic system are changed gradually its trajectories and attractors
typically change gradually, however, for certain parameter values sudden changes in the dynamic behavior
can occur. It is at this critical point that complicated spatio-temporal organization may occur (e.g. from a
steady-state to a limit cycle attractor). Close to bifurcations the system also becomes increasingly more
sensitive to parameter and initial condition changes. It is often proposed that bifurcations offer a selection
mechanism [Prigogine, 1985], as a dynamical system may respond very differently to very small changes in
their parameters.

15

https://www.studyc.info/
However, if the parameter space is divided by many bifurcations, the system becomes increasingly sensitive
to initial conditions and small parameter changes; in this sense its behavior becomes chaotic. It has been
argued that the most useful behavior lies instead in between full order and chaos. Langton [1990, 1992] has
shown (for one-dimensional cellular automata) that it is in this range of behavior that dynamical systems can
carry the most complicated “computations”. Computation here is used in a loose sense—not as the rate-
independent, symbolic manipulation of Turing-machines—meaning that information exchange between
elements of these systems is maximized in this range. In other words, Langton showed that the highest
correlation among the automata in a cellular lattice occur at this stage.

Kauffman [1993,] likewise hypothesized that “living systems exist in the [ordered] regime near the edge of
chaos, and natural selection achieves and sustains such a poised state”. This hypothesis is based on Packard’s
[1988] work showing that when natural selection algorithms are applied to dynamical systems, with the goal
of achieving higher discriminative power, the parameters are changed generally to lead these systems into
this transitional area between order and chaos. This idea is very intuitive, since chaotic dynamical systems
are too sensitive to parameter changes, that is, a single perturbation or mutation (structural perturbation) leads
the system into another completely different behavior (sensitive to damage). By contrast, ordered systems are
more resilient to damage, and a small parameter change will usually result in a small behavior change which
is ideal for smooth adaptation. However, even though very ordered systems can adapt by accumulation of
useful successful variations (because damage does not propagate widely), they may not be able ‘step out’ of
their particular organization in the presence of novel demands in their environment.

It is here that systems at the edge of chaos were thought to enter the scene; they are not as sensitive to damage
as chaotic systems, but still they are more sensitive than fully ordered systems. Thus, most mutations cause
only minor structural changes and can accumulate, while a few others may cause major changes in the
dynamics enabling a few dramatic changes in behavior. These characteristics of simultaneous mutation
buffering (to small changes) and dramatic alteration of behavior (in response to larger changes) is ideal for
evolvability [Conrad, 1983, 1990]. However, many of the real gene networks that have been successfully
modeled with dynamical systems (e.g. the network of segment polarity genes in Drosophila melanogaster
[Albert and Othmer, 2003]), exist in a very ordered regime, being very robust to structural changes [Chaves,
Albert and Sontag, 2005; Willadsen&Wiles, 2007; kauffman et al, 2003]. Still, other genetic regulatory
network models do operate close to criticality [Balleza et al, 2008]. It appears that evolution favors ordered,
very robust regimes of self-organization in gene networks – at least the ones involved in very conserved
regulatory pathways – though there is also evidence of near-critical regimes for increased evolvability.

Complex Self-organization
We have studied several computational systems said to be self-organizing in the sense described above. The
discrete logistic equation observes several ranges of ordered behavior according to its parameter r. For r #
3, the system converges to a single point steady state (independently of its initial value). For 3 # r # 4 the
system enters a series of bifurcations, meaning that it changes its attractor behavior, first from a steady-state
into a two-state limit cycle, and then progressively doubling the number of states in an attractor limit cycle
as r increases. Close to r = 4, the limit cycle becomes chaotic. That is, in the chaotic range, the slightest
change in the initial value, will lead to a completely different trajectory (though similarly chaotic). The
system goes from being independent to strongly dependent of initial conditions, though, in each range, the
attractor behavior of the equation is the same for random initial conditions. Thus, we can see the logistic
equation as self-organizing.

16

https://www.studyc.info/
But there is another aspect of the logistic equation that should be understood. In all of its ranges of behavior,
from full order to full chaos, the system is (fairly) reversible. That is, I can always obtain a specific initial
condition which caused some behavior, by formally running the system backwards. This means the system
is deterministic in both temporal directions. Formally, this means the state transition function is invertible.
(This is actually only true, if we decide to work on the lower half of its state space, since the logistic equation
is a quadratic function, it has always two possible solutions for the previous value of the current state, these
values are symmetric about the middle point of its state space). Some, resist calling this kind of reversible
systems self-organizing because they are not sufficiently complex. They reason that if a system is self-
organizing, when ran backwards it should be self-disorganizing, that is, it should lead to random initial
conditions, or to an incomplete knowledge of possible initial states. Indeed, complexity is typically equated
with the inability to describe the behavior of a system from the behavior of its components or predecessors.
This way, we ought to reserve the term self-organization to those irreversible systems whose behaviors must
be evaluated statistically. The logistic map shows “hints” of this backwards self-disorganization, but we can
still work out effectively its backwards trajectory to an initial condition by restricting the quadratic solutions
to half of its state space.

Random Boolean Networks are much more complicated than this. They are completely deterministic since
a certain state will always lead to the same next state (state-determinacy), however, we cannot usually know
exactly what the predecessor of a current state was. Systems like this are usually studied with statistical tools.
Even though the rules that dictate the next state of its components are simple and deterministic, the overall
behavior of the system is generally too complicated to predict and statistical analysis has to be performed.
For instance, Kauffman has shown that when K=2 (number of inputs to each node), his networks will have
on average basins of attraction with a length of states; if the output of one node is switched to the
other boolean value (perturbation), the trajectory returns to that cycle 85% of the time, while on the remaining
15% of the time it will “jump” into a different basin of attraction. Cellular automata (CA) fall into this same
category of deterministic, irreversible, self-organization.

Further Readings and References:

Albert, R., H.G. Othmer [2003]. “The topology of the regulatory interactions predicts the expression pattern of the
segment polarity genes in Drosophila melanogaster.” J Theor Biol. 223(1):1-18.
Balleza, E., E. R. Alvarez-Buylla, A. Chaos, S. Kauffman, I. Shmulevich, and M. Aldana. [2008]. “Critical Dynamics
in Genetic Regulatory Networks: Examples from Four Kingdoms.” PloS One 3 (6): e2456.
Chaves, M., R. Albert, and E.D. Sontag [2005]. “Robustness and fragility of Boolean models for genetic regulatory
networks.”J Theor Biol. 235(3):431-49.
Conrad. M. [1983], Adpatability. Plenum Press.
Conrad, M. [1990], “The geometry of evolutions”. In BioSystems Vol. 24, pp. 61-81.
Forrest, S. (Ed.) [1990]. Emergent Computation. MIT Press/ North-Holland. Special Issue of Physica D. Vol. 42.
Kauffman, S.A. [1993]. The Origins of Order: Self-Organization and Selection in Evolution. Oxford University Press.
Kauffman, Stuart, Carsten Peterson, Björn Samuelsson, and Carl Troein [2003]. “Random Boolean Network Models and
the Yeast Transcriptional Network.” Proceedings of the National Academy of Sciences of the United States of
America 100 (25): 14796–14799.
Klir, George J. [1991]. Facets of Systems Science. Plenum Press.
Langton, C. [1990]. “Computation at the edge of chaos: phase transitions and emergent computation”. In Forrest [1990]
pp12-37.
Langton, C. [1992], “Life at the edge of chaos”. In Artificial Life II. C. Langton (Ed.) Pp 41-91. Addison-Wesley
Packard, N. [1988] “Adaptation toward the edge of chaos”. In. Complexity in Biologic Modelling. S. Kelso and M.
Shlesinger.
Pattee, Howard H. [1978]."The complementarity principle in biological and social structures." In: Journal of Social and
Biological Structures Vol. 1, pp. 191-200.
Prigogine, I. [1985]. “New Perspectives on Complexity”. In: Sciences and Praxis of Complexity. pp.107-118.

17

https://www.studyc.info/
Willadsen, K. and Wiles, J. [2007] "Robustness and state-space structure of Boolean gene regulatory models". Journal
of Theoretical Biology, 249 (4), 749-765.
Wuensche, A., & Lesser, M. [1992]. The global dynamics of cellular automata: An atlas of basin of attraction fields of
one-dimensional cellular automata. Reading, MA: Addison-Wesley.

18

https://www.studyc.info/
5. Reality is Stranger than Fiction
Updated from a presentation in the “Biocomplexity” discussion section at the 9th European Conference on
Artificial Life, September 12, 2007 in Lisbon, Portugal

What can Artificial Life do about Advances in Biology?

“By extending the empirical foundation upon which biology is based beyond the carbon-chain life
that has evolved on Earth, Artificial Life can contribute to theoretical biology by locating life-as-we-
know-it within the larger picture of life-as-it-could-be”. [Langton, 1998, page 1]

From Langton’s original artificial life manifesto, the field was largely expected to free us from the confines
of “life-as-we-know-it” and its specific biochemistry. The idea of “life-as-it-could-be” gave us a scientific
methodology to consider and study the general principles of life at large. The main assumption of the field
was that instead of focusing on the carbon-based, living organization, life could be better explained by
synthesizing its “logical forms” from simple machines [Langton, 1989, page 11]—where, “fictional”
machines substituted real biochemistry. The expectation was that this “out-of-the-box”, synthetic
methodology would gain us a wider scientific understanding of life. We would be able to entertain alternative
scenarios for life, challenge the dogmas of biology, and ultimately discover the design principles of life.

Interestingly, during the 20 years since the first artificial life workshop, biology witnessed tremendous
advances in our understanding of life. True, biology operates at a completely different scale of funding and
in a much larger community base than artificial life (the impact factors of key journals in both fields differ
by an order of magnitude). But, still, it is from biology, not artificial life, that the strangest and most exciting
discoveries and design principles of life arise today. Consider looking at the [September 6, 2007] number of
Nature, with the quite apropos editorial title “Life as We Know it” [Vol. 449, 1], arguing for a comparative
genomics approach, with articles, for instance, moving towards evolutionary principles of gene duplication
[Wapinski et al, 2007]. Publications in the [September 2007 issue of] PLoS. Biol., also presented new
evidence towards updating or discovering general principles of life: for instance, Venter’s sequencing of his
diploid genome, which updates our expectations of differences in chromosome pairs [Levy et al, 2007]), and
the Ahituv et al [2007] study that challenges the idea that utraconserved DNA (across species) must be
functional. Since then, many advances, often enabled by big data approaches of computational biology, keep
being discovered; for instance, from large-scale comparative genomics, it has been found that retroviral
genomic sequences account for 6 to 14% of host genomes—-8% of human DNA is from endogenous
retroviruses, which comprises more DNA than the human proteome [Weiss & Stoye, 2013].

It is good to notice that this sort of work is not so much an exception, but has been a signature of research in
the biosciences in the last couple of decades. Consider cases such as the discovery of DNA transfer from
bacteria to the fly [Dunning Hotopp, 2007], extra-genomic inheritance in Arabidopsis [Lolle et al, 2005], or
the profound importance of non-coding RNA in life which is a major player in, among other features,
patterning [Martello et al, 2007] , essential gene regulation [Mattick, 2005], development [Mattick, 2007],
epigenetic neural development and modulation [Mehler & Mattick, 2007; Mattick & Mehler, 2008],
eukariotic complexity [Taft et al, 2007], etc. Moreover, advances such as these do not seem to be mere
epiphenomena of a specific life form. Indeed, they point at important organization principles—as those that
artificial life was supposed to provide. When we discover that non-transcribed RNA is involved in extra-
genomic inheritance or that most of the evolutionary innovation responsible for differences between
marsupials and placental mammals occurs in non-protein coding DNA [Mikkelsen et al, 2007], some
fundamental principles of the living organization are to be re-thought: the simple, generalized genotype-
phenotype mappings on which most of artificial life is based on, are just not enough to capture the principles

https://www.studyc.info/
of life as we know it. More intricate genomic structure, and its principles, need to be modeled and theories
need to be built to understand life.

One could go on and on about many other advances in biology. We can also point to themes at the forefront
of (bio)complexity theory that go largely overlooked in artificial life—though not completely (i.e. [Calabretta
et al, 2000; Hintze & Adami, 2007]). Perhaps the key topic in complexity theory today is that of modularity
in evolution [Schlosser & Wagner, 2004 ] and in networks [Newman, 2006; Guimerà et al 2007].
Nonetheless, looking at the papers accepted for the main sections of the latest Alife and ECAL conferences,
it is easy to see that most papers, not only do not discover or even address such issues, but largely trade in
biological and computational concepts that have not changed much since the field’s inception (see list of top
themes and terms in appendix). Is artificial life trapped in the (evolutionary) biology of twenty years ago?
Why is reality stranger and more surprising than fiction?

Clearly, there has been very widely successful artificial life research. First and foremost, artificial life has
been most successful as a means to study animal behavior, learning and cognition. Certainly, evolutionary
robotics and embodied cognition have had an impact in cognitive science. But is artificial life simply a better
way to do artificial intelligence? Moreover, one could argue that given the embodied nature of evolutionary
robotics, it would seem that it is bound to some kind of material reality, rather than synthesized by constituent
“logical forms” as Langton initially suggested.

But what to do about the organization of life itself? Surely the idea of explaining the living organization was
behind the origin of the field. For the purposes of this discussion, we must question ourselves why artificial
life does not produce more and surprising results about the living organization? Certainly, there is sound
research in the field with impact outside of it [e.g. Adami, 2006; Hintze & Adami, 2007]. But even the most
successful research in artificial life rarely goes beyond showing that artificial organisms can observe the same
behaviors as their real counterparts (i.e. selective pressures, epistasis, etc.). A problem for the field is that
as biotechnology gains more and more control of cellular processes, it is reasonable to ask what can one do
with artificial organisms that one cannot do with real bacteria? For instance, recent studies of the evolutionary
speed towards beneficial mutations were quite effectively done with E-coli [Perfeito et al, 2007], pointing
to a much larger rate of beneficial mutations in bacteria than previously thought, and shedding new light on
the general principal of clonal interference.

The point of this short statement is to discuss at this conference [ECAL 2007], how biocomplexity is dealt
within artificial life, twenty years after the field’s inception. Certainly the community can think of a variety
of responses to this lack of new principles of life coming out of research in artificial life—even in theoretical
biology. One concept that I venture may need updating in artificial life is its view of the genotype/phenotype
relationship. Langton proposed that we generalize this relationship, but this meant that research in the field
largely regarded the two as indistinguishable. While this move at fist glance seems appropriate to deal with
the complexity of genomic-proteomic interaction, it prevents us from studying the specific roles each plays
in the living organization. Genotype and phenotype are intertwined in a complex manner, but each operates
under different principles that are often overlooked in artificial life. Thus, artificial life rarely approaches
issues of genomic structure and regulation, or the co-existence of DNA and RNA as different types of
informational carriers. This could well be because artificial life models seem to trade most often on the
concept of Mendelian gene than on the molecular biology gene. In other words, artificial life models tend to
regard genes solely as mechanisms of generational (vertical) inheritance, rather than as (informational)
mechanisms of ontogenetic (horizontal) development, regulation, maintenance, phenotypic plasticity, and
response to environmental change. This way, most artificial life models do not test, or even deal with,
possible genomic structure architectures and their impact on development and evolution. This is a big

20

https://www.studyc.info/
shortcoming in the field since, as we have seen in the last two decades, the molecular biology gene and the
genomic structure it implies are behind many essential principles of life—from hypersomatic mutation in
vertebrate immunity to speciation.

Additionally, it is most often the case that artificial organisms in artificial life models are designed with many
top-down features, rather than emerging out of artificial biochemical machines. For instance, typically the
genes of artificial organisms encode pre-defined computer operations. Not only is the encoding pre-defined,
but the function of individual genes is also pre-programmed, rather than emergent from some artificial
chemistry—what is typically emergent is the behavior of a collection of such “atomic” genes and genotypes.

It is interesting to note that when biologists were looking for the location of genetic information for
inheritance, they naturally assumed that it would reside in proteins. They knew of DNA chemically, but its
sheer inertness deemed it unfit for the job. It took some time to realize that relative inertness was really the
point--- from Griffith’s experiment in 1928 to Avery’s in 1944, the implications of which were only fully
accepted much later , probably costing Avery a deserved Nobel [Judson, 2003]. This episode illustrates how
reality very often surprises the best scientific expectations of the day—a big problem for Artificial Life, as
long as it defines itself as the study of life-as-it-could-be, since it implies a science built on what scientists
think life is and not on what experiments show it is. For instance, the biochemical difference between highly
inert memory molecules and highly reactive, functional ones, while often overlooked in artificial life as a
design principle, is ultimately the hallmark of life [Rocha and Hordijk, 2005; Brenner, 2012]. Indeed,
Venter’s achievement in successfully replicating a living cell with a “prosthetic genome” until the original
organism’s phenotype is fully re-programmed (see chapter 1), should lead Artificial Life scientists to ponder
at least the question of what is it about life’s design principle that makes it easier to synthesize a working
prosthetic genome than a working “prosthetic” proteome or metabolome? Perhaps, Langton’s view of
artificial life being built-up from simple machines, may have clouded the fact that life as we know it is made
of biochemical constituents with very different chemical and functional roles: chiefly, DNA (long-term,
random-access memory), RNA (short-term memory and symbol processing) and proteins (functional
machines). Perhaps more attention should be directed to the “logical forms” of these lower level, structural
constituents that produce life, before we can tackle “life-as-it-could-be”.

ACKNOWLEDGMENTS
I am extremely grateful to Karola Stotz at Indiana University for exciting conversations that directly lead to
this text. I also thank Janet Wiles for performing the Leximancer analysis, and Peter Todd for patiently
listening as some of the ideas here discussed were germinating.

Further Readings and References

Ahituv, N., et al. [2007]. “Deletion of Ultraconserved Elements Yields Viable Mice”. PLoS Biology Vol. 5,
No. 9, e234.

Brenner, S. [2012]. “Turing centenary: Life’s code script.” Nature 482 (7386): 461-461.

Calabretta, R., Nolfi, S., Parisi, D. and Wagner, G.P. [2000]. “Duplication of modules facilitates the evolution
of functional specialization”. Artificial Life, 6 , 69-84.

Hintze, A. and C. Adami [2007]. “Evolution of complex modular biological networks”. arXiv.org:0705.4674.

21

https://www.studyc.info/
Dunning Hotopp, J. C., et al. [2007] "Widespread Lateral Gene Transfer from Intracellular Bacteria to
Multicellular Eukaryotes," Science doi:10.1126.

Guimerà, R, Sales-Pardo, M, Amaral, LAN. [2007] “Classes of complex networks defined by role-to-role
connectivity profiles”. Nature Phys. 3, 63-69.

Judson, H.F. [2003]. "No Nobel Prize for Whining". New York Times, October 20, 2003.

Levy, S. et al. [2007] PLoS Biol. 5, e254.

Lolle, S. J., et al [2005]. “Genome-wide non-mendelian inheritance of extra-genomic information in

Arabidopsis”, Nature 434, 505-509.

Mikkelsen, T.S. et al [2007]. “Genome of the marsupial Monodelphis domestica reveals innovation in
non-coding sequences”. Nature 447, 167-177. ,

Martello, G. Et al [2007]. “MicroRNA control of Nodal signalling” Nature. doi:10.1038/nature06100;

Mattick, J.S. [2005] “The functional genomics of noncoding RNA”. Science. 309 (5740): 1570-3.

Mattick JS.[2007]. “A new paradigm for developmental biology”. J Exp Biol. ;210(Pt 9):1526-47

Mattick, JS & MF Mehler [2008] “RNA editing, DNA recoding and the evolution of human cognition”.
Trends in Neurosciences, 31 (5):227-233.

Mehler MF, Mattick JS. [2007]. “Noncoding RNAs and RNA editing in brain development, functional
diversification, and neurological disease”. Physiol Rev. 87(3):799-823

Newman, M.E.J. [2006]. “Modularity and community structure in networks”. PNAS. 103: 8577.

.Perfeito L, L. Fernandes, C. Mota and I. Gordo [2007]. “Adaptive Mutations in Bacteria: High Rate and
Small Effects”. Science. 317 (5839):813 - 815.

Rocha, L.M. and W. Hordijk [2005]. "Material Representations: From the Genetic Code to the Evolution of
Cellular Automata". Artificial Life. 11 (1-2):189 - 214

Schlosser, G. and G. P. Wagner (Eds.) [2004]. Modularity in Development and Evolution.

Taft RJ, Pheasant M, Mattick JS. [2007].”The relationship between non-protein-coding DNA and eukaryotic
complexity”. Bioessays. 29(3):288-99.

Wapinski, I., A. Pfeffer, N. Friedman, and A. Regev [2007]. "Natural history and evolutionary principles of
gene duplication in fungi". Nature 449, 54-61.1.

Weiss & Stoye [2013]. “Our Viral Inheritance.” Science.340 (6134): 820-821

22

https://www.studyc.info/
APPENDIX:

Top themes extracted from all abstracts accepted to ECAL 2007, produced the
Leximancer (courtesy of Janet Wiles)

23

https://www.studyc.info/
Top Themes produced from Leximancer set at 65% coverage themes (courtesy of Janet
Wiles)

Top co-occurring (stemmed) word pairs in abstracts

neural--network
chang--environ
artifici--life
simul--result
autonom--robot
evolutionari--algorithm
evolutionari--robot
comput--simul
genet--algorithm
robot--mobil
cellular--automata
interact--between
artifici--chemistri
agent--adapt
pressur--select
neural--control

24

https://www.studyc.info/
6. Von Neumann and Natural Selection
“Turing invented the stored-program computer, and von Neumann showed that the description is
separate from the universal constructor. This is not trivial. Physicist Erwin Schrödinger confused the
program and the constructor in his 1944 book What is Life?, in which he saw chromosomes as
“architect's plan and builder's craft in one”. This is wrong. The code script contains only a description
of the executive function, not the function itself.” [Brenner, 2012]

6.1 Von Neumann’s Self-Reproduction Scheme

Von Neumann thought of his logical model of self-
reproduction as an answer to the observation that,
unlike machines, biological organisms have the
ability to self-replicate while seemingly increasing
their complexity without limit. Mechanical
artefacts are instead produced via more
complicated factories (as opposed to self-
production) and can only degenerate in their
complexity. He was searching for a complexity
threshold beyond which systems may self-
reproduce (with no outside control) while possibly
increasing their complexity.

Von Neumann concluded that this threshold entails

a memory-stored description Φ(X) that can be
interpreted by a universal constructor automaton
A to produce any automaton X; if a description of
A, Φ(A), is fed to A itself, then a new copy of A is
obtained. However, to avoid a logical paradox of self-reference, the description, which cannot describe itself,
must be both copied (uninterpreted role) and translated (interpreted role) into the described automaton. This
way, in addition to the universal constructor, an automaton B capable of copying any description, Φ(X), is
included in the self-replication scheme. A third automaton C is also included to perform all the manipulation
of descriptions necessary—a sort of operating system. To sum it up, the self-replicating system contains the
set of automata (A + B + C) and a description Φ(A + B + C); the description is fed to B which copies it three
times (assuming destruction of the original); one of these copies is then fed to A which produces another
automaton (A + B + C); the second copy is then handled separately to the new automaton which together with
this description is also able to self-reproduce; the third copy is kept so that the self-reproducing capability
may be maintained (it is also assumed that A destroys utilized descriptions). Notice that the description, or
program, is used in two different ways: it is both translated and copied. In the first role, it controls the
construction of an automaton by causing a sequence of activities (active role of description). In the second
role, it is simply copied (passive role of description). In other words, the interpreted description controls
construction, and the uninterpreted description is copied separately, passing along its stored information
(memory) to the next generation. This parallels the horizontal and vertical transmission of genetic
information in biological organisms, which is all the more remarkable since Von Neumann proposed this
scheme before the structure of the DNA molecule was uncovered by Watson and Crick[1953]—though after
the Avery-MacLeod-McCarty [1944] experiment which identified DNA has the carrier of genetic
information.

25

https://www.studyc.info/
 “The concept of the gene as a symbolic representation of the organism–a code script–is a funda-
mental feature of the living world and must form the kernel of biological theory.” [Brenner, 2012]

The notion of description-based self-reproduction implies a language. A description must be cast on some
symbol system while it must also be implemented by some physical or a logical structure. When A interprets
a description to construct some automaton, a semantic code is utilized to map instructions into construction
commands to be performed. When B copies a description, only its syntactic aspects are replicated. Now, the
language of this semantic code presupposes a set of primitives (e.g. parts and processes) for which the
instructions are said to “stand for”. Descriptions are not universal insofar as they refer to these building blocks
which cannot be changed without altering the significance of the descriptions. The building blocks ultimately
produce the dynamics, behavior, and/or functionality of the overall system, and may be material or
computational. In Biology, we can think of the genetic code as instantiating such a language. Genes are
descriptions that encode specific parts: amino acids chains. In a computational setting, parts are typically
logical operations, but they can also be, for example, the building blocks of neural networks coded by genetic
algorithms and L-Systems. Von Neumann [1966] (posthumously aided by Arthur Burks) produced a
specification of a universal constructor using a 29-state cellular automaton. Implementations of this
automaton appeared only fairly recently [e.g. Pesavento, 1995, see Sipper, 1998]

6.2 Open-ended evolution and natural selection

“Biologists ask only three questions of a living organism: how does it work? How is it built? And
how did it get that way? They are problems embodied in the classical fields of physiology,
embryology and evolution. And at the core of everything are the tapes containing the descriptions
to build these special Turing machines.” [Brenner, 2012]

Perhaps the most important consequence of separate descriptions in Von Neumann’s self-reproduction
scheme (and Turing’s Tape) is its opening the possibility for open-ended evolution [Rocha, 1998; McMullin,
2000]. As Von Neumann [1966] discussed, if the description of the self-reproducing automata is changed
(mutated), in a way as to not affect the basic functioning of (A + B + C) then, the new automaton (A + B +
C)` will be slightly different from its parent. Von Neumann used a new automaton D to be included in the
self-replicating organism, whose function does not disturb the basic performance of (A + B + C); if there is
a mutation in the D part of the description, say D`, then the system (A + B + C + D) + Φ(A + B + C + D`)
will produce (A + B + C + D`) + Φ(A + B + C + D`). Von Neumann [1966, page 86] further proposed that
non-trivial self-reproduction should include this “ability to undergo inheritable mutations as well as the ability
to make another organism like the original”, to distinguish it from “naive” self-reproduction like growing
crystals.

Notice that changes in (A + B + C + D) are not heritable, only changes in the description, Φ(A + B + C + D),
are inherited by the automaton’s offspring and are thus relevant for evolution. This ability to transmit
mutations (vertically) is precisely at the core of the principle of natural selection of modern Darwinism.
Through variation (mutation) populations of different organisms are produced; the statistical bias these
mutations impose on reproduction rates of organisms will create survival differentials (fitness) on the
population which define natural selection. In principle, if the language of description is rich enough, an
endless variety of organisms can be evolved: open-ended evolution.

The evolvability of a self-reproducing system is dependent on the parts used by the semantic code. If the parts
are very simple, then the descriptions will have to be very complicated, whereas if the parts possess rich
dynamic properties, the descriptions can be simpler since they will take for granted a lot of the dynamics that

26

https://www.studyc.info/
otherwise would have to be specified. In the genetic system, genes do not have to specify the functional
characteristics of the proteins produced, but simply the string of amino acids that will produce that
functionality “for free” [Moreno et al, 1994]. Furthermore, there is a trade-off between programmability and
evolvability [Conrad, 1983, 1990] which grants some self-reproducing systems no evolutionary potential
whatsoever. When descriptions require high programmability they will be very sensitive to damage. Low
programmability grants self-reproducing systems the ability to change without destroying their own
organization, though it also reduces the space of possible evolvable configurations [Rocha, 2001].

Turing and Von Neumann were the first to correctly formalize the required inheritance mechanism behind
neo-Darwinian evolution by Natural Selection. This understanding of the most fundamental design principle
of life, puts Turing and Von Neumann on the Parthenon of great thinkers in Biology, alongside Darwin and
Mendel. The dovetailing of computational thinking and biology, inherent in the cybernetics movement of
Turing, Von Neumann, Shannon, Wiener and others, emphasizes how (material) control of symbolic
information is the hallmark of both computation and biocomplexity.

Further Readings and References:

Avery, Oswald T.; Colin M. MacLeod, Maclyn McCarty [1944]. "Studies on the Chemical Nature of the Substance
Inducing Transformation of Pneumococcal Types: Induction of Transformation by a Desoxyribonucleic Acid
Fraction Isolated from Pneumococcus Type III". Journal of Experimental Medicine 79 (2): 137–158.
Brenner, S. [2012]. “Turing centenary: Life’s code script.” Nature 482 (7386): 461-461.
Conrad. M. [1983], Adpatability. Plenum Press.
Conrad, M. [1990], “The geometry of evolutions”. In BioSystems Vol. 24, pp. 61-81.
McMullin, B. [2000]. “John von Neumann and the Evolutionary Growth of Complexity: Looking Backwards, Looking
Forwards”. Artificial Life 6(4):347-361.
Moreno, A., A. Etxeberria, and J. Umerez [1994], “Universality Without Matter?”. In Artificial Life IV, R. Brooks and
P. Maes (Eds). MIT Press. pp 406-410
Pesavento, U. [1995] An implementation of von Neumann's self-reproducing machine. Artificial Life 2(4):337-354.
Rocha, L.M. [1998]."Selected Self-Organization and the Semiotics of Evolutionary Systems". In: Evolutionary Systems:
The Biological and Epistemological Perspectives on Selection and Self- Organization, . S. Salthe, G. Van de
Vijver, and M. Delpos (eds.). Kluwer, pp. 341-358.[2]
Rocha, L.M. [2001]. "Evolution with material symbol systems". Biosystems. 60, pp. 95-121
Sipper, M. [1998]. “Fifty Years of Research on Self-replication: an Overview”. Artificial Life, 4 (3), pp. 237-257.
von Neumann, John [1966]. The Theory of Self-Reproducing Automata. Arthur Burks (Ed.) University of Illinois Press.
Watson Jd, Crick Fh [1953]. "Molecular structure of nucleic acids; a structure for deoxyribose nucleic acid". Nature 171
(4356): 737–8.

27

https://www.studyc.info/
7. Modeling Evolution: Evolutionary Computation
“How does evolution produce increasingly fit organisms in environments which are highly uncertain for
individual organisms? How does an organism use its experience to modify its behavior in beneficial ways (i.e.
how does it learn or ‘adapt under sensory guidance’)? How can computers be programmed so that problem-
solving capabilities are built up by specifying ‘what is to be done’ rather than ‘how to do it’?” [Holland, 1975,
page 1]

These were some of the questions concerning John Holland when he thought of Genetic Algorithms (GA’s)
in the 1960's. All these questions were shown to be reducible to a problem of optimizing multi-parameter
functions. Nature’s “problem” is to create organisms that reproduce more (are more fit) in a particular
environment: the environment-organism coupling dictates the selective pressures, and the solutions to these
pressures are organisms themselves. In the language of optimization, the solutions to a particular problem
(say, an engineering problem), will be selected according to how well they solve that problem. GA’s are
inspired by natural selection as the solutions to our problem are not algebraically calculated, but rather found
by a population of solution alternatives which is altered in each time step of the algorithm in order to increase
the probability of having better solutions in the population. In other words, GA’s, or other Evolutionary
Strategies (ES) such as Evolutionary Programming (EP), explore the multi-parameter space of solution
alternatives for a particular problem, by means of a population of encoded strings (standing for alternatives)
which undergo variation (crossover and mutation) and are reproduced in a way as to lead the population to
ever more promising regions of this search space (selection) [Goldeberg, 1989; Mitchell, 1999; De Jong,
2006].

7.1 Evolutionary Strategies and Self-Organization

The underlying idea of computational ES is the separation of solutions for a particular problem (e.g. a
machine) from descriptions of those solutions (memory). GA’s work on these descriptions and not on the
solutions themselves, that is, variation is applied to descriptions, while the respective solutions are evaluated,
and the whole (description-solution) selected according to this evaluation. Such machine/description
separation follows von Neumann’s self-reproducing scheme (see chapter 6) which is able to increase the
complexity of the (organization of) machines described. Therefore, the form of organization evolved by GA’s
is not self-organizing in the sense of a boolean network or cellular automata (see chapter 4). Even though the
solutions are obtained from the interaction of a population of elements, and in this sense following the general
rules usually observed by computationally emergent systems (e.g. Langton [1988], Mitchell [1992]), they
do not self-organize since they rely on the selective pressures of some environment (in ES, defined by an
explicit or implicit fitness function). The order so attained is not a result of the internal dynamics of a
collection of interacting elements, but is instead dictated by the external selection criteria. In this sense, ES
follow an organizing scheme that is driven by external selection of encoded symbolic descriptions (a “Turing
tape” ). It is perhaps useful to think that ES are modeling the most fundamental design principle of biological
systems: natural selection. While self-organizing systems model the dynamical characteristics of matter, ES
model the existence of, external, selective pressures on populations of symbolic descriptions of some system.
While self-organization models material dynamics, ES models the selection of information about dynamics.

7.2 Development and morphogenesis: self-organization and selection come together

Since the original introduction of GA’s, many subsequent developments had to do with the inclusion of a
developmental stage, or intermediate layers between genotype and phenotype; in other words, the creation
of some artificial morphogenesis or regulation. The idea has been to encode rules that will themselves self-

28

https://www.studyc.info/
organize to produce a phenotype, rather than the direct encoding of the phenotype itself, or the introduction
of gene regulation . As discussed in class, these rules often use L-System grammars which dictate production
system programs [Wilson, 1988] leading to some phenotype. The most important advantage of this
intermediate stage, as explored by Kitano [1990], Gruau [1993], Belew [1992] and others, is the ability to
code for much larger structures than a direct encoding allows. In practical terms, they have solved some of
the scalability problems of encoding (e.g.) neural networks in GA’s, by reducing the search space
dramatically.

L-system grammars are higher-level descriptions of self-organizing developmental processes. However, these
first approaches used solely context-free, state-determined, L-System grammars, compromising epistasis (or
mutual, non-linear, influence of genetic descriptions amongst each other) in the simulation of self-organizing
development. Dellaert and Beer [1994] and Kitano [1994], for instance, used Boolean networks to simulate
genetic epistasis and self-organization. In other words, the GA encodes rules which construct Boolean
networks whose nodes stand for aspects of the phenotypes we wish to evolve on some physical simulation.
In Dellaert and Beer’s model, the nodes stand for cell mitosis and other characteristics. This way, the
solutions of the GA are self-organizing systems whose attractor behavior dictates pre-defined phenotypic
traits.

These approaches in effect offer an emergent morphology, that is, they encode rules which will themselves
self-organize into some phenotype (instead of strict programming of morphology). The indirect encoding
further allows the search to occur in a reduced space, amplified through development. An interesting side
effect has to do with the appearance of modularity traits on the evolved phenotypes [Wagner, 1995].
Subsequent developments paid even more attention to the contextual regulation that indirect encodings afford
to the search [Rocha 1995, 1997]. More recently, given our expanded view of genomics, other intermediate
layers between genotype and phenotype have been explored, such as transcription regulation [Reil, 1999;
Hallinan & Wiles, 2004] and RNA Editing [Rocha et al, 2006]. The inclusion of more sophisticated regulation
of genetic information prior to translation, while not necessarily including a self-organizing component,
allows us to model a much more realistic genotype/phenotype/environment interaction. Instead of genotypes
used exclusively for Mendelian inheritance (see chapter 5) of (directly encoded) phenotypic traits, ES with
genotype regulation allow us to model the contextual, plastic development of phenotypes we have come to
understand via modern Genomics—thus also learning additional design principles for bio-inspired
computation [Huang et al, 2007].

The most important aspect of GA’s with emergent morphologies is the utilization in the same model of an
external selection engine (the GA) coupled to a particular self-organizing dynamics (e.g. Boolean networks)
standing for some materiality. Such schemes bring together, computationally, the two most important aspects
of evolutionary systems: self-organization and selection. These models belong to a category of self-
organization referred to as Selected Self-Organization which is based on symbolic memory [Rocha, 1996,
1997, 1998]. Selected Self-Organization with distributed memory is also possible in autocatalytic structures,
though its evolutionary potential is much smaller than the local memory kind [Rocha, 2001][Vasas, 2010].
The reason lies in Von Neumann’s notion of self-reproduction (see chapter 6). The introduction of symbolic
descriptions allows a much more sophisticated form of communication: structures are constructed from static
descriptions and do not have to reproduce through some complicated, and limited process of self-inspection.
In other words, separate descriptions can be used to reliably construct any kind of structure in an open-ended
manner, while self-inspection relies on only those structures that happen to be able to make copies of
themselves. As an example, a non-genetic protein-based life form, would have to rely only on those proteins
that could make direct copies of themselves [Rocha, 2001].

29

https://www.studyc.info/
Further Readings and References:
Belew, R.K. [1993]."Interposing an Ontogenic Model Between Genetic Algorithms and Neural Networks." In:
Advances in neural information processing (NIPS5). J. Cowan (Ed.). Morgan Kaufmann.
De Jong, K.A.[2006]. Evolutionary Computation: A Unified Approach. MIT Press.
Dellaert, F. and R.D. Beer [1994]."Toward an evolvable model of development for autonomous agent synthesis." In:
Artificial Life IV: Proceedings of the Fourth International Workshop on the Synthesis and Simulation of Living
Systems. R. Brooks and P. Maes (Eds.). MIT Press.
Goldberg, D. E. [1989]. Genetic Algorithms in Search, Optimization, and Machine Learning. Addison-Wesley.
Gruau, Frédéric [1993]."Genetic Sythesis of Modular Neural Networks." In: Proceedings of the fifth international
conference on Genetic Algorithms. S. Forrest. Morgan Kauffman, pp 318-325.
Hallinan, J. And J Wiles [2004]. “Asynchronous Dynamics of an Artificial Genetic Regulatory Network”. 9th
International Conference on Artificial Life. MIT Press.
Holland, John H. [1975]. Adaptation in Natural and Artificial Systems. University of Michigan Press.
C. Huang, J. Kaur, A. Maguitman, L.M. Rocha[2007].""Agent-Based Model of Genotype Editing". Evolutionary
Computation, 15(3).
Kitano, H. [1990]."Designing Networks using Genetic Algorithms with Graph Generation System." In: Complex
Systems Vol. 4, pp 461-476.
Kitano, Hiroaki [1994]."Evolution of Metabolism for Morphogenesis." In: Artificial Life IV: proceedings of the fourth
international workshop on the synthesis and simulation of living systems. R. Brooks and P. Maes (Eds.). MIT
Press.
Mitchell, M. [1992]. “Genetic algorithms”. In Lectures in Complex Systems. L. Nadel and D. Stein (Eds.). SFI Studies
in the Science of Complexity Vol. V, Addison-Wesley. pp 3-87.
Mitchell, M. [1999]. An Introduction to Genetic Algorithms. MIT Press.
Reil, T. 1999. “Dynamics of Gene Expression in an Artificial Genome - Implications for Biological and Artificial
Ontogeny”. In Proceedings of the 5th European Conference on Advances in Artificial Life (September 13 - 17,
1999). D. Floreano, J. Nicoud, and F. Mondada, Eds. Lecture Notes In Computer Science, vol. 1674.
Springer-Verlag, London, 457-466
Rocha, Luis M. [1995]."Contextual Genetic Algorithms: Evolving Developmental Rules." In: Advances in Artificial
Life. F. Moran, A. Moreno, J.J. Merelo, and P. Chacon (Eds.). Series: Lecture Notes in Artificial Intelligence,
Springer-Verlag. pp. 368-382.
Rocha, Luis M. [1996]."Eigenbehavior and symbols." In: Systems Research Vol. 12, No. 3, pp. 371-384.
Rocha, Luis M. [1997]. Evidence Sets and Contextual Genetic Algorithms: Exploring Uncertainty, Context, and
Embodiment in Cognitive and Biological Systems. PhD. Dissertation. SUNY Binghamton.
Rocha, Luis M. [1998]." Selected Self-Organization and the Semiotics of Evolutionary Systems." In: Evolutionary
Systems: The Biological and Epistemological Perspectives on Selection and Self- Organization. S. Salthe, G.
Van de Vijver, and M. Delpos (eds.). Kluwer Academic Publishers, pp. 341-358.
Rocha, Luis M. [2001]. "Evolution with Material Symbol Systems". Biosystems. Vol. 60, pp. 95-121.
Rocha, L.M., A. Maguitman, C. Huang, J. Kaur, and S. Narayanan. [2006].""An Evolutionary Model of Genotype
Editing". In: Artificial Life 10: Tenth International Conference on the Simulation and Synthesis of Living
Systems. L.M.Rocha, L. Yaeger, M. Bedau, D. Floreano, R. Goldstone, and A. Vespignani (Eds.). MIT Press,
pp. 105-111.
Vasas, Vera, Eors Szathmary, and Mauro Santos [2010]. " Lack of evolvability in self-sustaining autocatalytic networks:
A constraint on the metabolism-first path to the origin of life. ". Proceedings of the National Academy of
Sciences of the United States of America: 0912628107.
Wagner, Gunter [1995] “Adaptation and the modular design of organisms”. In: Advances in Artificial Life. F. Moran,
A. Moreno, J.J. Merelo, and P. Chacon (Eds.). Series: Lecture Notes in Artificial Intelligence, Springer-
Verlag. pp. 317-328.

30

https://www.studyc.info/
An Introduction to Evolutionary Computation

Talib S. Hussain

Department of Computing and Information Science

Queen’s University, Kingston, Ont. K7L 3N6

[email protected]

https://www.studyc.info/
 Abstract possible in a fixed amount of time. For a search space
with only a small number of possible solutions, all the
Researchers in many fields are faced with
solutions can be examined in a reasonable amount of
computational problems in which a great number of
time and the optimal one found. This exhaustive
solutions are possible and finding an optimal or even a
search, however, quickly becomes impractical as the
sufficiently good one is difficult. A variety of search
search space grows in size. Traditional search
techniques have been developed for exploring such
algorithms randomly sample (e.g., random walk) or
problem spaces, and a promising approach has been the
heuristically sample (e.g., gradient descent) the search
use of algorithms based upon the principles of natural
space one solution at a time in the hopes of finding the
evolution. This tutorial will introduce the basic
optimal solution. The key aspect distinguishing an
principles underlying most evolutionary algorithms, as
evolutionary search algorithm from such traditional
well as some of the key details of the four most popular
algorithms is that it is population-based. Through the
methods: genetic algorithms, genetic programming,
adaptation of successive generations of a large number
evolutionary strategies, and evolutionary programming.
of individuals, an evolutionary algorithm performs an
The aim of the tutorial is to introduce the participants to
efficient directed search. Evolutionary search is
the jargon and principles of the field of evolutionary
generally better than random search and is not
computation, and to encourage the participants to
susceptible to the hill-climbing behaviors of gradient-
consider the potential of applying evolutionary
based search.
optimization techniques in their own research.
1. Introduction 2. Basic Evolutionary Computation
In an evolutionary algorithm, a representation
An important area in current research is the
scheme is chosen by the researcher to define the set of
development and application of search techniques based
solutions that form the search space for the algorithm.
upon the principles of natural evolution. Most readers,
A number of individual solutions are created to form an
through the popular literature and typical Western
initial population. The following steps are then
educational experience, are probably aware of the basic
repeated iteratively until a solution has been found
concepts of evolution. In particular, the principle of the
which satisfies a pre-defined termination criterion.
‘survival of the fittest’ proposed by Charles Darwin
Each individual is evaluated using a fitness function that
(1859) has especially captured the popular imagination.
is specific to the problem being solved. Based upon
We shall use this as a starting point in introducing
their fitness values, a number of individuals are chosen
evolutionary computation.
to be parents. New individuals, or offspring, are
The theory of natural selection proposes that the
produced from those parents using reproduction
plants and animals that exist today are the result of
operators. The fitness values of those offspring are
millions of years of adaptation to the demands of the
determined. Finally, survivors are selected from the old
environment. At any given time, a number of different
population and the offspring to form the new population
organisms may co-exist and compete for the same
of the next generation.
resources in an ecosystem. The organisms that are most
The mechanisms determining which and how many
capable of acquiring resources and successfully
parents to select, how many offspring to create, and
procreating are the ones whose descendants will tend to
which individuals will survive into the next generation
be numerous in the future. Organisms that are less
together represent a selection method. Many different
capable, for whatever reason, will tend to have few or
selection methods have been proposed in the literature,
no descendants in the future. The former are said to be
and they vary in complexity. Typically, though, most
more fit than the latter, and the distinguishing
selection methods ensure that the population of each
characteristics that caused the former to be more fit are
generation is the same size.
said to be selected for over the characteristics of the
The remainder of the paper presents the traditional
latter. Over time, the entire population of the ecosystem
definitions of the four most common evolutionary
is said to evolve to contain organisms that, on average,
algorithms: genetic algorithms (Holland, 1975), genetic
are more fit than those of previous generations of the
programming (Koza, 1992, 1994), evolutionary
population because they exhibit more of those
strategies (Rechenberg, 1973), and evolutionary
characteristics that tend to promote survival.
programming (Fogel et al., 1966). The traditional
Evolutionary computation techniques abstract these
differences between the approaches involve the nature
evolutionary principles into algorithms that may be used
of the representation schemes, the reproduction
to search for optimal solutions to a problem. In a search
operators, and the selection methods.
algorithm, a number of possible solutions to a problem
are available and the task is to find the best solution 3. Genetic Algorithms

https://www.studyc.info/
 The most popular technique in evolutionary 4. Genetic Programming
computation research has been the genetic algorithm.
An increasingly popular technique is that of genetic
In the traditional genetic algorithm, the representation
programming. In a standard genetic program, the
used is a fixed-length bit string. Each position in the
representation used is a variable-sized tree of functions
string is assumed to represent a particular feature of an
and values. Each leaf in the tree is a label from an
individual, and the value stored in that position
available set of value labels. Each internal node in the
represents how that feature is expressed in the solution.
tree is label from an available set of function labels.
Usually, the string is “evaluated as a collection of
The entire tree corresponds to a single function that may
structural features of a solution that have little or no
be evaluated. Typically, the tree is evaluated in a left-
interactions” (Angeline, 1996, p. 4). The analogy may
most depth-first manner. A leaf is evaluated as the
be drawn directly to genes in biological organisms.
corresponding value. A function is evaluated using as
Each gene represents an entity that is structurally
arguments the result of the evaluation of its children.
independent of other genes.
Genetic algorithms and genetic programming are
similar in most other respects, except that the
a) 1 1 0 1 0 0 c) 1 1 0 1 0 1 reproduction operators are tailored to a tree
representation. The most commonly used operator is
subtree crossover, in which an entire subtree is swapped
b) 1 0 1 0 0 1 d) 1 0 1 0 0 0 between two parents (see Figure 3). In a standard
genetic program, all values and functions are assumed
Crossover Point to return the same type, although functions may vary in
Figure 1: Bit-String Crossover of Parents a & b the number of arguments they take. This closure
to form Offspring c & d principle (Koza, 1994) allows any subtree to be
considered structurally on par with any other subtree,
and ensures that operators such as sub-tree crossover
will always produce legal offspring.
a) 1 0 1 0 0 1 b) 1 0 1 0 1 1
+ +
Figure 2: Bit-Flipping Mutation of Parent a
to form Offspring b
− 1 − +
The main reproduction operator used is bit-string
crossover, in which two strings are used as parents and
new individuals are formed by swapping a sub-sequence 3 2 3 2 1 4
between the two strings (see Figure 1). Another popular (a) (c)
operator is bit-flipping mutation, in which a single bit in
× ×
the string is flipped to form a new offspring string (see
Figure 2). A variety of other operators have also been
developed, but are used less frequently (e.g., inversion,
5 + 5 1
in which a subsequence in the bit string is reversed). A
primary distinction that may be made between the
various operators is whether or not they introduce any
1 4
new information into the population. Crossover, for (d)
(b)
example, does not while mutation does. All operators
are also constrained to manipulate the string in a
Figure 3: Subtree Crossover of Parents a & b
manner consistent with the structural interpretation of
to form Offspring c & d
genes. For example, two genes at the same location on
two strings may be swapped between parents, but not 5. Evolutionary Strategies
combined based on their values.
In evolutionary strategies, the representation used
Traditionally, individuals are selected to be parents
is a fixed-length real-valued vector. As with the bit-
probabilistically based upon their fitness values, and the
strings of genetic algorithms, each position in the vector
offspring that are created replace the parents. For
corresponds to a feature of the individual. However,
example, if N parents are selected, then N offspring are
the features are considered to be behavioral rather than
generated which replace the parents in the next
structural. “Consequently, arbitrary non-linear
generation.
interactions between features during evaluation are

https://www.studyc.info/
expected which forces a more holistic approach to A typical selection method is to select all the
evolving solutions” (Angeline, 1996, p. 4). individuals in the population to be the N parents, to
The main reproduction operator in evolutionary mutate each parent to form N offspring, and to
strategies is Gaussian mutation, in which a random probabilistically select, based upon fitness, N survivors
value from a Gaussian distribution is added to each from the total 2N individuals to form the next
element of an individual’s vector to create a new generation.
offspring (see Figure 4). Another operator that is used is
intermediate recombination, in which the vectors of two 7. Current Issues
parents are averaged together, element by element, to In current research, the line distinguishing these
form a new offspring (see Figure 5). The effects of different approaches has started to blur. Researchers in
these operators reflect the behavioral as opposed to each technique have begun to examine more complex
structural interpretation of the representation since representation schemes and to apply a variety of
knowledge of the values of vector elements is used to selection methods. Many genetic algorithm researchers
derive new vector elements. are examining the use variable-length representations
and analyzing how such representations grow in size
over the course of evolution (Wu & Lindsay, 1996).
a) 1.3 0.4 1.8 0.2 0.0 1.0 b) 1.2 0.7 1.6 0.2 0.1 1.2 Many genetic algorithms now use selection methods,
such as elitist recombination, in which parents compete
Figure 4: Gaussian Mutation of Parent a
with their offspring for survival into the next generation
to form Offspring b
(Thierens, 1997). Some genetic programming
researchers have begun to examine the effects of
allowing multiple types of functions and values into the
a) 1.2 0.3 0.0 1.7 0.8 1.2 representation. The benefits of such strongly typed
c) 1.0 0.4 0.5 1.4 0.5 1.2 genetic programming are only beginning to be explored
b) 0.8 0.5 1.0 1.1 0.2 1.2 (Haynes et al., 1996).
Figure 5: Intermediate Recombination of Parents a & b References
to form Offspring c Angeline, P.J. (1996) “Genetic programming’s continued evolution,”
Chapter 1 in K.E. Kinnear, Jr. and P.J. Angeline (Eds.), Advances
The selection of parents to form offspring is less in Genetic Programming 2. Cambridge, MA: MIT Press, p. 1 -20.
Darwin, C. (1859) On the Origin of Species by Means of Natural
constrained than it is in genetic algorithms and genetic Selection. London: John Murray.
programming. For instance, due to the nature of the Fogel, L.J., Owens, A.J. & Walsh, M.J. (1966) Artificial Intelligence
through Simulated Evolution. New York: John Wiley.
representation, it is easy to average vectors from many Haynes, T.D., Schoenefeld, D.A. & Wainwright, R.L. (1996) “Type
individuals to form a single offspring. In a typical inheritance in strongly typed genetic programming,” Chapter 18 in
K.E. Kinnear, Jr. and P.J. Angeline (Eds.), Advances in Genetic
evolutionary strategy, N parents are selected uniformly Programming 2. Cambridge, MA: MIT Press. p. 359-376.
randomly (i.e., not based upon fitness), more than N Holland, J.H. (1975) Adaptation in Natural and Artificial Systems.
Ann Arbor, MI: University of Michigan Press.
offspring are generated through the use of Koza, J.R. (1992) Genetic Programming: On the Programming of
recombination, and then N survivors are selected Computers by Means of Natural Selection. Cambridge, MA: MIT
Press.
deterministically. The survivors are chosen either from Koza, J.R. (1994) Genetic Programming II: Automatic Discovery of
the best N offspring (i.e., no parents survive) or from Reusable Programs. Cambridge, MA: MIT Press.
the best N parents and offspring (Spears et al., 1993). Rechenberg, I. (1973) Evolutionsstrategie: Optimierung Technischer
Systeme nach Prinzipien der Biologischen Evolution. Stuttgart:
Frommann-Holzboog Verlag.
6. Evolutionary Programming Spears, W.M., DeJong, K.A., Back, T., Fogel, D.B., & deGaris, H.
(1993) “An overview of evolutionary computation,” Proceedings
The representations used in evolutionary of the 1993 European Conference on Machine Learning.
programming are typically tailored to the problem Thierens, D. (1997) “Selection schemes, elitist recombination, and
selection intensity,” Proceedings of the Seventh International
domain (Spears et al., 1993). One representation Conference on Genetic Algorithms. San Francisco, CA: Morgan
commonly used is a fixed-length real-valued vector. Kauffman, p. 152-159.
Wu, A. & Lindsay, R.K. (1996) “A survey of intron research in
The primary difference between evolutionary genetics,” In Voigt, H., Ebelin, W., Rechenberg, I., & Schwefel,
programming and the previous approaches is that no H. (Eds.), Parallel Problem Solving from Nature - PPSN IV.
Berlin: Springer-Verlag, p. 101-110.
exchange of material between individuals in the
population is made. Thus, only mutation operators are
used. For real-valued vector representations,
evolutionary programming is very similar to
evolutionary strategies without recombination.

https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/
Evolutionary Systems
All biological systems result from an evolutionary process. The sophistication, robustness, and
adaptability of biological systems represent a powerful motivation for replicating the mechanisms
of natural evolution in the attempt to generate software and hardware systems with characteristics
comparable to those of biological systems. More than 40 years ago, computer scientists and
engineers began developing algorithms inspired by natural evolution (Rechenberg 1965; Fogel et
al. 1966; Holland 1975) to generate solutions to problems that were too difficult to tackle with
other analytical methods. Evolutionary computation rapidly became a major field of machine
learning and system optimization and, more recently, it spread into the area of hardware design by
exploiting new technologies in reconfigurable electronic circuits, computer-assisted
manufacturing, material production tech nologies, and robotics. Before delving into the features
of natural and artificial evolution, we wish to emphasize that there is a major, and often neglected,
difference between these two processes. Whereas natural evolution does not have a predefined
goal and is essentially an open-ended adaptation process, artificial evolution is an optimization
process that attempts to find solutions to predefined problems. Therefore, while in natural
evolution the fitness of an individual is defined by its reproductive success (number of offspring),
in artificial evolution the fitness of an individual is a function that measures how well that
individual solves a predefined problem. The consequence of this difference is that artificial
evolution, as it is formulated today, cannot possibly hope to match the diversity and creativity
generated by natural evolution because, by definition, artificially evolved systems will all tend to
satisfy the predefined problem.

https://www.studyc.info/
https://www.studyc.info/
Again, evolutionary adaptation does not necessarily imply progress in the two meanings of the
word described earlier. Natural evolution may simply increase diversity by continuously
generating new organisms that occupy new environmental niches. Or, it may increase complexity
by incrementally adding new features to previous ones, provided that previous features do not
represent a cost for the organism, do not interact negatively with new features, or simply have a
higher probability to be preserved than to be replaced by the error-prone copy mechanism.
Considering the enormous explanatory power and relative simplicity of the basic tenets of
evolutionary theory, we might expect to find in the literature a compact and universal model that
formally describes the evolution of populations, something akin to the laws of thermodynamics or
to Newton’s laws of physics. In practice, the complexity of the factors that affect the mechanisms
and dynamics of evolution has not yet been sufficiently understood to allow the development of a
universal formalism. Nonetheless, several formal models have been developed to address specific
issues, mainly in the field of population genetics. It is worth pointing out that the great majority of
these formal models describe evolutionary phenomena in terms of their ef fect on the variation rate
of the population size or of a given character of the evolving individuals. In other words, formal
measures of evolution, if we may liberally call them so, describe frequencies of the occurrence of
given characters, or of given types of organism, over generations. For example, these models
predict that in a relatively stable environment the percentage of individuals with fitter
characteristics will gradually grow until they dominate the population (Fisher 1930). These models
do not address the notion of performance and progress in evolving populations, but only the change
in proportion of organisms of a certain type.

https://www.studyc.info/
The Genotype
So far, we have not yet explained how individual characters can be inherited and
modified. In 1865 Mendel arrived at the conclusion that individuals reproduce by
transmitting specific particles, now known as genetic material, to their own offspring.
Recent progress in genetics (the discipline studying the structure and behavior of
genes) and in functional genomics (the discipline studying the role of genes in
organisms) has provided several clues to the molecular mechanisms and processes that
support inheritance and variation. Although Darwin was probably not aware of
Mendel’s conclusions when he formulated the theory of evolution, genetics has
become an integral part of modern evolutionary theories.

https://www.studyc.info/
https://www.studyc.info/
https://www.studyc.info/