Analyzing fault tolerance on parallel genetic programming by means of dynamic-size Populations

This paper presents an experimental research on the size of individuals when dynamic size populations are employed with Genetic Programming (GP). By analyzing the individual's size evolution, some ideas are presented for reducing the length of the best individual while also improving the quality. This research has been performed studying both individual's size and quality of solutions, considering the fixed-size populations and also dynamic size by means of the plague operator. We propose an improvement to the Plague operator, that we have called Random Plague, that positively affects the quality of solutions and also influences the individuals' size. The results are then considered from a quite different point of view, the presence of processors failures when parallel execution over distributed computing environments are employed. We show that results strongly encourage the use of Parallel GP on non fault-tolerant computing resources: experiments shows the fault tolerant nature of Parallel GP.