Abstract:

Spiking neuron has precise timing characteristics of multispiking coding information.According to this feature, a new supervised learning algorithm based on convolution calculation for multilayer spiking neural networks is proposed.The algorithm uses the convolution calculation of the kernel functions to convert the discrete spiking sequence to a continuous function;in the network architecture consisting of a multilayer feedforward network of spiking neurons,the gradient method based on kernel function convolution is adopted to obtain learning rules and adjust the synaptic weights of neuron connections.In the experiment,effectiveness of the algorithm for learning spiking sequence is verified,and then the algorithm is applied to classify the Iris data set.The results show that the algorithm enables the learning of complex spatiotemporal patterns of spiking sequence,and it has high classification accuracy for nonlinear pattern classification problem

Key words: spiking neural networks;supervised learning;convolution calculation;gradient descent