Holland J.H. Adaptation in Natural and Artificial Systems. An Introductory Analysis with Applications to Biology, Control, and Artificial Intelligence

MIT Press, 1992. — 225 p.The first technical descriptions and definitions of adaptation come from biology. In that context adaptation designates any process whereby a structure is progressively modified to give better performance in its environment. The structures may range from a protein molecule to a horse's foot or a human brain or, even, to an interacting group of organisms such as the wildlife of the African veldt. Defined more generally, adaptive processes have a critical role in fields as diverse as psychology ("learning"), economics ("optimal planning"), control, artificial intelligence , computational mathematics and sampling ("statistical inference"). Basically, adaptive processes are optimization proces ses, but it is difficult to subject them to unified study because the structures being modified are complex and their performance is uncertain. Frequently non-additive interaction (i.e., "epistasis" or "nonlinearity") makes it impossible to determine the performance of a structure from a study of its isolated parts. Moreover possibilities for improved performance must usually be exploited at the same time that the search for further improvements is pressed. While these difficulties pose a real problem for the analyst, we know that they are routinely handled by biological adaptive processes, qua process e.s The approach of this book is to set up a mathematical framework which makes it possible to extract and generalize critical factors of the biological processes. Two of the most important generalizations are: (1) the concept of a schema as a generalization of an interacting, coadapted set of genes, and (2) the generalization of genetic operators such as crossing-over, inversion, and mutation. The schema concept makes it possible to dissect and analyze complex "nonlinear" or "epistatic" interactions, while the generalized genetic operators extend the analysis to studies of learning, optimal planning, etc. The possibility of "intrinsic parallelism" - the testing of many schemata by testing a single structure - is a direct offshoot of this approach. The book develops an extensive study of intrinsically parallel processes and illustrates their uses over the full range of adaptive processes, both as hypotheses and as algorithms.
The book is written on the assumption that the reader has a familiarity with probability and combinatorics at the level of a first course in finite mathematical structures, plus enough familiarity with the concept of a system to make the notion of "state" a comfortable working tool . Readers so prepared should probably read the book in the given order, moving rapidly (on the first reading) over any example or proof offering more than minor difficulties. A good deal of meaning can still be extracted by those with less mathematics if they are willing to abide the notation, treating the symbols (with the help of the Glossary) as abbreviations for familiar intuitive concepts.The General Setting
A Formal Framework
Illustrations
Schemata
The Optimal Allocation of Trials
Reproductive Plans and Genetic Operators
The Robustness of Genetic Plans
Adaptation of Codings and Representations
An Overview
Interim and Prospectus