Swarm Intelligence
Table of Contents
Part I
Chapter 1
Chapter 2
Chapter 3
Chapter 4
Chapter 5
Chapter 6
Part I lays the
groundwork for our journey into the world of particle swarms and swarm
intelligence that occurs later in the book. We visit big topics such as
life, intelligence, optimization, adaptation, simulation and modeling.
BACK
Chapter 1 – Models and Concepts of Life and Intelligence first looks at what kinds of phenomena can be included under these terms. What is life? This is an important question of our historical era, as there are many ambiguous cases. Can life be created by man? What is the role of adaptation in life and thought? And why do so many natural adaptive systems seem to rely on randomness?
Is cultural evolution Darwinian? – Some think so; the question of
evolution in culture is central to this volume. The Game of Life and
cellular automata in general are computational examples of emergence, which
seems to be fundamental to life and intelligence, and some artificial life
paradigms are introduced. The chapter begins to inquire about the nature of
intelligence and reviews some of the ways that researchers have tried to
model human thought. We conclude that intelligence just means "the qualities
of a good mind," which of course might not be defined the same by everybody.
BACK
Chapter 2 – Symbols, Connections, and Optimization by Trial and Error
is
intended to provide a background that will make the later chapters
meaningful. What is optimization and what does it have to do with minds? We
describe aspects of complex fitness landscapes, and some methods that are
used to find optimal regions on them. Minds can be thought of as points in
high-dimensional space: what would be needed to optimize them? Symbols as
discrete packages of meaning are contrasted to the connectionist approach
where meaning is distributed across a network. Some issues are discussed
having to do with numeric representations of cognitive variables and
mathematical problems.
BACK
Chapter 3 -- On Our Nonexistence as Entities: The Social Organism considers the various zoom-angles that can be used to look at living and thinking things. Though we tend to think of ourselves as autonomous beings, we can be considered as macro-entities hosting multitudes of cellular or even sub-cellular guests, or as micro-entities inhabiting a planet that is alive. The chapter addresses some issues about social behavior. Why do animals live in groups? How do the social insects manage to build arches, organize cemeteries, stack wood chips? How do bird flocks and fish schools stay together? And what in the world could any of this have to do with human intelligence? (Hint: it has a lot to do with it.)
Some interesting questions have had to be answered before robots could do
anything on their own. Rodney Brooks’ subsumption architecture builds
apparently goal-directed behavior out of modules. And what’s the difference
between a simulated robot and an agent? Finally, Chapter 3 looks at computer
programs that can converse with people. How do they do it? –Usually by
exploiting the shallowness or mindlessness of most conversation.
BACK
Chapter 4 – Evolutionary Computation Theory and Paradigms describes the
four major computational paradigms that use evolutionary theory for problem
solving in some detail. The fitness of potential problem solutions is
calculated, and the survival of the fittest allows better solutions to
reproduce. These powerful methods are known as the "second best way" to
solve any problem.
BACK
Chapter 5 – Humans – Actual, Imagined, and Implied starts off musing on
language as a bottom-up phenomenon. The chapter goes on to reviews the
downfall of behavioristic psychology and the rise of cognitivism –
meanwhile, social psychology kept simmering in the background. Clearly there
is a relationship between culture and mind, and a number of researchers have
tried to write computer programs based on that relationship. As we review
various paradigms, it becomes apparent that a lot of people think that
culture must be similar to Darwinistic evolution. Are they the same? How are
they different?
BACK
Chapter 6 – Thinking is Social. This chapter eases us into our own
research on social models of optimization. The Adaptive Culture Model is
based on Axelrod’s Culture Model – in fact it is exactly like it except for
one little thing: individuals imitate their neighbors, not on the basis of
similarity, but on the basis of their performance. If your neighbor has a
better solution to the problem than you do, you try to be more like them. It
is a very simple algorithm with big implications.
BACK
Part II: Particle Swarm Optimization and Collective Intelligence
Part II focuses on our particle swarms paradigm, and the collective
and individual intelligence that arises within the swarm. We first introduce
the conceptually simplest version of particle swarms: binary particle
swarms, and then discuss the "workhorse" of particle swarms, the real-valued
version. Variations on the basic algorithm and the performance of the
particle swarm on benchmark functions precede a review of a few
applications.
BACK
Chapter 7- The Particle Swarm begins by suggesting that the same simple processes that underlie cultural adaptation can be incorporated into a computational paradigm. Multivariate decision making is reflected in a binary particle swarm. The performance of binary particle swarms is then evaluated on a number of benchmarks.
The chapter then describes the real-valued particle swarm optimization
paradigm. Individuals are depicted as points in a shared high-dimensional
space. The influence of each individual’s successes and those of neighbors
is similar to the binary version, but change is now portrayed as movement
rather than probability. The chapter concludes with a description of the use
of particle swarm optimization to find the weights in a simple neural
network.
BACK
Chapter 8 – Variations and Comparisons is a somewhat more technical look
at what various researchers have done with the basic particle swarm
algorithm. We first look at the effects of the algorithm’s main parameters,
and at a couple of techniques for improving performance. Are particle swarms
actually just another kind of evolutionary algorithm? There are reasons to
think so, and reasons not to. Considering the similarities and differences
between evolution and culture can help us understand the algorithm and
possible things to try with it.
BACK
Chapter 9 – Applications reviews a few of the applications of particle
swarm optimization. The use of particle swarm optimization to evolve
artificial neural networks is presented first. Evolutionary computation
techniques have most commonly been used to evolve neural network weights,
but have sometimes been used to evolve neural network structure or the
neural network learning algorithm. The strengths and weaknesses of these
approaches are reviewed. The use of particle swarm optimization to replace
the learning algorithm and evolve both the weights and structure of a neural
network is described. An added benefit of this approach is that it makes
scaling or normalization of input data unnecessary. The classification of
the Iris Data Set is used to illustrate the approach. Although a feedforward
neural network is used as the example, the methodology is valid for
practically any type of network.
BACK
Chapter 10 – Implications and Speculations reviews the implications of
particle swarms for theorizing about psychology and computation. If social
interaction provides the algorithm for optimizing minds, then what must that
be like for the individual? Various social- and computer-science
perspectives are brought to bear on the subject.
BACK
Chapter 11 – And In Conclusion… looks back at some of the motifs that
were woven through the narrative.
BACK
Appendix
A – Statistics for Swarmers is where we review some methods for
scientific experimental design and data analysis. The discussion is a
high-level overview to help researchers design their investigations; you
should be conversant with these tools if you’re going to evaluate what you
are doing with particle swarm optimization – or any other stochastic
optimization, for that matter. Included are sections on descriptive and
inferential statistics, confidence intervals, Student’s t-test,
one-way analysis of variance, factorial and multivariate ANOVA, regression
analysis, and the chi-square test of independence. The material in this
appendix provides you with sufficient information to perform some of the
simple statistical analyses. In more complex areas, we provide mainly
descriptive material, and point you to a couple of good (in our opinion)
statistics texts.
BACK