Supervised Learning of Single-Layer Spiking Neural Networks for Image Classification

Abstract

The traditional artificial neural networks encode information through the spike firing rate. Spiking neural networks fall into the third-generation artificial neural network models, which use the precisely timed spike trains to encode neural information. The computational models can accurately simulate the neural network activities of human brain, and provide powerful capabilities of signal processing to solve the complex problem. In this paper, we propose a supervised learning algorithm for single-layer spiking neural networks based on the spike train kernel function, which can implement the complex spatio-temporal pattern learning of spike trains. Furthermore, a pattern classifier based on single-layer spiking neural networks is constructed for image recognition problem. We test the learning performance of the proposed algorithm by the image classification task on the LabelMe dataset. The experimental results show that the proposed algorithm has got good image classification accuracy for the test dataset, and the different sizes of receptive fields influence classification accuracies significantly.