A Supervised Learning Algorithm to Binary Classification Problem for Spiking Neural Networks

Abstract

Spiking neural networks (SNN) are known as the third generation neural network, which can simulate biological neural networks signals and has stronger computing power. In contrast to the model classification tasks previously mentioned in machine learning, the Tempotron algorithm is a biologically rational and temporal coding supervised synaptic learning rule that enables neurons to efficiently learn a wide range of decision rules. Embedding information in the space-time structure of spikes rather than simply the average spike emission frequency. In this paper, we adopt Tempotron algorithm to perform binary classification task on the imported Fashion MNIST dataset and adopt gradient descent algorithm to update the synaptic weight during the training process. The two conditions of sending spikes and no sending spikes are taken as the classification standard. The experimental results show that this method has high learning accuracy and efficiency can classify the dataset accurately, and solve complex and real-time problems better.