A new supervised learning algorithm with the adaptive decay time for the spike neural network

The paper proposes a new supervised learning algorithm for spiking neural networks (SNNs). This algorithm is based on the error back-propagation algorithms. However, it is considered that the decay time of the spike response function is a variable that needs to be adjusted during network training instead of synaptic weights. By calculating and adjusting the decay time variable, the spike firing of neurons in the network will ensure that the output neurons fire spikes at the desired time. The accuracy and efficiency of the proposed algorithm are considered through the XOR classification problem. The results show that the proposed algorithm has a successful rate of classification equal to the Multi-Spikeprop algorithm and higher than the Spikeprop algorithm, whereas the number of epochs is smaller.