A Supervised Learning Rule for Recurrent Spiking Neural Networks with Weighted Spikes

Abstract

As a brain-inspired artificial neural network computational model, a recurrent spiking neural network is composed of biologically plausible spiking neurons, which has taken on increasing importance in this study field mainly include complex network structure and implicit nonlinear mechanism. The paper presents a learning rule with spike weight for recurrent spiking neural networks, allowing for real-time communication system of complex spatiotemporal spike trains simulating organisms. First, a context layer with connectivity through copying the hidden layer between the input layer and the output layer is provided. The total error of the network for learning spike train patterns is then introduced, as well as a rule of spike weight based on different phases of spikes. Furthermore, the proposed supervised learning rule defines the learning process for synaptic weights in all layers based on the power-up of weighted spikes to transmit more information. In addition, the learning algorithm has been successfully tested and evaluated for major factors, such as different lengths and frequencies for desired output spike trains, demonstrating that high precision learning is possible even with limited iterative resources. Finally, an analysis of the recurrent layer parameters is conducted, including neuron number and connectivity degree.