Parallel Programs Are More Evolvable than Sequential Programs

This paper presents a novel phenomenon of the Genetic Parallel Programming (GPP) paradigm - the GPP accelerating phenomenon. GPP is a novel Linear Genetic Programming representation for evolving parallel programs running on a Multi-ALU Processor (MAP). We carried out a series of experiments on GPP with different number of ALUs. We observed that parallel programs are more evolvable than sequential programs. For example, in the Fibonacci sequence regression experiment, evolving a 1-ALU sequential program requires 51 times on average of the computational effort of an 8-ALU parallel program. This paper presents three benchmark problems to show that the GPP can accelerate evolution of parallel programs. Due to the accelerating evolution phenomenon of GPP over sequential program evolution, we could increase the normal GP’s evolution efficiency by evolving a parallel program by GPP and if there is a need, the evolved parallel program can be translated into a sequential program so that it can run on conventional hardware.