Evolution of Typed Expressions describing Artificial Nervous Systems

Evolutionary algorithms implement adaptation or learning in analogy to natural selection over a population of individuals competing in a certain environment. Similar to the genetic programming paradigm introduced by J. Koza [6] who uses LISP-S-expressions our structures undergoing adaptation are hierarchical, typed expressions (terms).1 The set of generable structures is the set of all possible compositions of typed terms that can be composed recursively from a problem specific set of function symbols ℱ = {f1, f2,… fN} with arities A = {a1, … ,aN}, ai = (minargfi,maxargfi). For each function fi the arity can vary between a minimum and maximum number of arguments. According type descriptions Ƭ = {t1 ,… , tN} define the expressions’ type structures, where the ti are regular expressions over ℱ ∪ S and S represents a set of standard types.2