Comparing Evolutionary Strategy Algorithms for Training Spiking Neural Networks

Abstract

Spiking Neural Networks are considered as the third gen- eration of Articial Neural Networks, these neural networks naturally process spatio-temporal information. Spiking Neural Networks have been used in several elds and application areas; pattern recognition among them. For dealing with supervised pattern recognition task a gradient- descent based learning rule (Spike-prop) has been developed, however it has some problems like no convergence. To overcome these prob- lems, metaheuristic algorithms such as Evolutionary Strategy have been used. In this work, three variants of the Evolutionary Strategy algorithm are compared for training Spiking Neural Networks. Several well-known benchmark dataset are used to test the capabilities of the algorithms.