Session:

WORLDS - Regeneration and Learning in Developmental Systems

Title:

Growing and Evolving Bitstreams in a Fixed Configuration Space

Authors:

R. Timothy Edwards

Abstract:

An indirect-mapping approach to evolvable algorithms is presented that is specifically intended for evolvable hardware. The conventional wisdom is that because hardware is configured by downloading a bitstream of fixed size and hardware-specific format, hardware is not suited to genetic algorithms in which the size of the chromosome may vary. The algorithm presented herein departs from the traditional representation of a genome as the configuration bitstream of the hardware itself. Instead, the genome represents a set of rules from which the configuration bitstream is built, in a process analogous to biological growth, or ontogeny. Chromosomes are generated largely through a process of self-organization that determines both the length and content. Evolved circuits grow to a state of maturity, and demonstrate an effective model of hardware self-repair.