A framework for evolution of computer systems on the fly

摘要

Hardware devices that automatically adapt to changes in environmental conditions and to defects in physical structure have long been studied. The new field of evolvable hardware has focused on combining artificial evolutionary mechanisms, inspired by nature, and reprogrammable devices to realize real robust systems. This paper proposes a framework for "on the fly" evolution of a computer system. The framework has two subsystems running in parallel - one dedicated to task execution and the otherreserved to continuously improve computer system performance. The evolution framework is in the phylogenetic-epigenetic plane, where chromosomes are able to learn. The goal is to provide the system with open-ended possibilities, as in nature. Algorithmsand framework are defined and initial results presented.