Sub-machine-code Genetic Programming

CPUs are often seen as sequential, however they have a high degree of internal parallelism, typically operating on 32 or 64 bits simultaneously. This paper explores the idea of exploiting this internal parallelism to extend the scope of genetic programming (GP) and improve its eeciency. We call the resulting form of GP sub-machine-code GP. The diierences between sub-machine-code GP and the usual form of GP are purely semantic and largely language independent, i.e. any GP system can potentially be used to do sub-machine code GP. In this chapter this form of GP and some of its applications are presented. The speed up obtained with this technique on Boolean classiication problems is nearly 2 orders of magnitude.