A discussion on training algorithm for timing of multiple spikes in Spiking Neural Networks

— Automatic adjustment of the number of units —

Abstract

Extended SpikeProp, which is a kind of spiking neural networks, can learn the timing of spikes. Our research group has discussed the method to learn not only the spike timings but also the number of spikes on extended SpikeProp. But, the proposed method is sensitive to the number of hidden units. In this article, we discuss the method to adjust the number of hidden units. Our method improved the rate of successful learning from 80% to 95% in simple experiments.