仿生型脉冲神经网络学习算法和网络模型

Bionic learning algorithm for spiking neuron networks and network model

为解决脉冲神经网络训练困难的问题,基于仿生学思路,提出脉冲神经网络的权值学习算法和结构学习算法,设计一种含有卷积结构的脉冲神经网络模型,搭建适合脉冲神经网络的软件仿真平台。实验结果表明,权值学习算法训练的网络对MNIST数据集识别准确率能够达到84.12%,具备良好的快速收敛能力和低功耗特点;结构学习算法能够自动生成网络结构,具有高度生物相似性。

To solve the difficult problem of training spiking neuron networks,the weight learning algorithm and structure lear-ning algorithm of spiking neuron networks were proposed based on the idea of bionics.A spiking neuron network model with convolution structure was designed,and a software simulation platform suitable for the spiking neuron network was built.Experimental results show that the network trained using the weight learning algorithm can identify the MNIST data set with an accuracy rate of 84.12%,and it has good learning ability of fast convergence and low power consumption.The structure learning algorithm can automatically generate network structure and it has high biological similarity.