Unsupervised Image Classification with Adversarial Synapse Spiking Neural Networks

Abstract

In this paper, we propose a spiking neural network architecture capable of learning different features between different labels. Our architecture, referred to as adversarial synapse spiking neural networks, pays more attention to the details of positive samples by inputting pairs of positive samples and negative samples. The proposed method uses a pair of adversarial input synapses, excitatory input synapses and inhibitory input synapses. Spike-timing dependent plasticity is used to train the weights in excitatory input synapses. Results shows the proposed method can significantly improve the image classification results of spiking neural networks.