STDP and Competition Learning in Spiking Neural Networks and its application to Image Classification

Spiking neural networks (SNNs), regarded as the third generation artificial neural networks (ANNs), can well explain the behavior of biological neurons. Recently, the research on the application of spiking neural networks has attracted much attention, especially in the image recognition field. To solve the problem of ANNs' lack of biological rationality, this paper combines Spike Timing Dependent Plasticity (STDP) with competitive learning to realize the MNIST dataset classification. A simple two-layer network structure, which includes an input layer and a processing layer is adopted. With the MNIST dataset as input, spike trains are generated based on frequency coding. A competitive learning mechanism is adopted in the processing layer to train the network, while during the learning and training process, we adopted the STDP power-law learning rule to update weights to achieve unsupervised learning image classification, and the classification accuracy reaches 83.179%. The results show that the network proposed in this paper achieves good performance, fast training speed and more biological rationality.