带延迟调整的脉冲神经元梯度下降学习算法  

Gradient Descent Learning Algorithm for Spiking Neuron with Delay Adjustment

作　　者：杨静 徐彦[2] 赵欣 YANG Jing;XU Yan;ZHAO Xin(School of Management,Zhuhai Campus,Beijing Normal University,Zhuhai,Guangdong 519087,China;College of Information Science and Technology,Nanjing Agricultural University,Nanjing 210095,China)

机构地区：[1]北京师范大学珠海分校管理学院,广东珠海519087 [2]南京农业大学信息科技学院,南京210095

出　　处：《计算机工程》2019年第9期169-175,共7页Computer Engineering

基　　金：国家自然科学基金(61503031);广东省科技计划基金(2016A040403029);广东省哲学社会科学“十二五”规划学科共建项目(GD15XGL02)

摘　　要：脉冲神经元有监督学习算法通过梯度下降法调整神经元的突触权值,但目标学习序列长度的增加会降低其精度并延长学习周期。为此,提出一种带延迟调整的梯度下降学习算法。将每个突触的延迟作为学习参数,在学习过程中调整权值,同时对突触的延迟时间进行梯度下降调整,从而使神经元激发出目标脉冲序列。实验结果表明,该算法在不增加算法复杂度的情况下,能够提高神经元学习复杂脉冲序列的能力,且收敛速度较快。The spiking neuron supervised learning algorithm adjusts the synaptic weight of the neuron by gradient descent method,but the accuracy gets low and the learning period gets long as the length of the target learning sequence increases.Therefore,a gradient descent learning algorithm with delay adjustment is proposed.The delay of each synapse is used as a learning parameter to adjust the weight during the learning process,and at the same time,the gradient delay adjustment of the synaptic delay time is performed to make neurons trigger a target spiking sequence.Experimental results show that the proposed algorithm can improve the ability of neurons to learn complex spiking sequences without increasing the complexity of the algorithm,and the convergence speed is faster.

关 键 词：脉冲神经元 脉冲序列 梯度下降 神经元突触 延迟学习 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]