Surpress Redundant Spikes in SpikeProp-Locally Application of Removing Redundant Delays

SpikeProp, which is proposed by Booij, is a kind of spiking neural networks. It can learn the timing of output spikes, but cannot adjust the number of output spikes. To enable SpikeProp to perform time series signal processing, our research group has discussed a learning algorithm for SpikeProp without redundant output spikes. The method consists of two techniques: adaptive weight decay (AWD) and removing redundant delays (RD). As a part of these researches, we discuss a method to locally application of removing redundant time delays in this article. Since AWD works differently on each part of network, RD should not be applied uniformity. By simple experiments, we showed that only RD between input layer and hidden layer is enough for performance.