Population management for automatic design of algorithms through evolution

Describes a population-based search in the ADATE (Automatic Design of Algorithms Through Evolution) system, which maintains chains of gradually bigger and better programs. The main challenge is to avoid missing links that lead to entrapment in local optima. To avoid entrapment, the ADATE system employs iterative re-expansion of programs, population maintenance using a syntactic complexity/evaluation value-ordering scheme and three different diversification methods that strive to avoid too similar programs. When combined with general program transformations, these techniques enable ADATE to synthesize recursive programs with automatic construction of recursive help functions. We also briefly present experimental results supporting the proposition that the automatic synthesis of complex programs from “first principles” is indeed possible, but only if vast computational resources are employed effectively