A cooperative method for supervised learning in Spiking neural networks

Abstract

In Spiking neural networks, information is encoded in separate spike times. The traditional gradient descent based learning algorithm (SpikeProp) trends to be trapped in local optima and cannot converge if the negative synaptic weights are allowed. In this paper, a cooperative PSO (Particle Swarm Optimization) method is proposed for its supervised learning. A simplified neural network structure is suggested. The CPSO-based learning method can improve both the weights of the spike neurons and the delays between the neurons. Both the positive and negative weights can be preserved by the biological neurons. Experiments on benchmark problems show the proposal is reliable and efficient for learning spike patterns.