A Modular Approach to Construction of Spiking Neural Networks

Abstract

In this paper, we propose a modular approach to construction of multi-layer spiking neural networks with a Coulomb energy function based learning algorithm for training each module. In this approach, a single-layer spiking neural network is constructed and trained with the Coulomb energy function based learning algorithm. If the learning result is not sufficiently good, another layer is added, and the input is the output of the previous layer. The process continues until a desired learning result is achieved. The approach eliminates the need for advance determination of the number of hidden layers and the need for error-backpropagation training in multi-layer spiking neural networks. Experimental results of classifying a two-ring-shaped dataset and segmenting an aerial image show that our proposed modular multi-layer spiking neural network requires a simple learning algorithm and achieves better results compared with other approaches.