A Framework for Evolving Spiking Neural P Systems

Abstract

In current literature, there is a lack of research on the optimization of spiking neural P systems (SN P systems) and, consequently, also a lack of automation to do this process of optimization. We address this gap by designing a genetic algorithm (GA) framework that transforms an initial SN P system Πinit, designed to approximate a function f(w,x, y, . . .) = z, into a smaller or more precise system Πfinal that also approximates the output z given the same input/s w,x, y, . . .. The design of the GA framework is constrained by evolving Πinit only through its topology. The rules inside the neurons must stay constant, while the synapses and neurons may vary. The results of the experiments conducted show that evolving the topology of a designed Πinit using genetic algorithms does not only lessen its number of neurons and synapses, but also helps it achieve a higher precision. The GA framework is especially effective on Πinit’s containing the subgraph of an already better SN P system that computes f.