Explorations in Parallel Linear Genetic Programming

Linear Genetic Programming (LGP) is a powerful problem-solving technique, but one with several significant weaknesses. LGP programs consist of a linear sequence of instructions, where each instruction may reuse previously computed results. This structure makes LGP programs compact and powerful, however it also introduces the problem of instruction dependencies. The notion of instruction dependencies expresses the concept that certain instructions rely on other instructions. Instruction dependencies are often disrupted during crossover or mutation when one or more instructions undergo modification. This disruption can cause disproportionately large changes in program output resulting in non-viable offspring and poor algorithm performance. Motivated by biological inspiration and the issue of code disruption, we develop a new form of LGP called Parallel LGP (PLGP). PLGP programs consist of n lists of instructions. These lists are executed in parallel, and the resulting vectors are summed to produce the overall program output. PLGP limits the disruptive effects of crossover and mutation, which allows PLGP to significantly outperform regular LGP. We examine the PLGP architecture and determine that large PLGP programs can be slow to converge. To improve the convergence time of large PLGP programs we develop a new form of PLGP called Cooperative Coevolution PLGP (CC PLGP). CC PLGP adapts the concept of cooperative coevolution to the PLGP architecture. CC PLGP optimizes all program components in parallel, allowing CC PLGP to converge significantly faster than conventional PLGP. We examine the CC PLGP architecture and determine that performance