多突触连接脉冲神经元的突触延迟在线监督学习算法

打开文本图片集

关键词：脉冲神经网络；在线监督学习；突触延迟学习；多突触连接中图分类号：TP183 文献标志码：A 文章编号：1001-3695（2025）08-023-2421-07doi：10.19734/j.issn.1001-3695.2024.11.0496

Online supervised learning algorithm for synaptic delays of spiking neurons with multiple synaptic connections

Wang Xiangwen 1，2† ，Zou Li³，Fan Jingxing1 （1.CollegeofforationineringsMalUersityHu7O;2CgefputerecE ginering，stlUsito；ralelicalical Equipment Center，Lanzhou 730020，China）

Abstract：Neuroscience studies haveshown thatsynapticdelay playsa positiverole in neural information procesing，and multiple synapticconnectionsare widelydistributed inthenervous system.However，mostof thecurrnt spiking neuralnetworksare modeled withasinglesynapticconnectionmode，andtheinfluenceof synapticdelayisnotfullconsidered in the designofsupervised learningalgorithms，whichlimitstheirpotentialperformance.Thispaperconstructedanetworkof spiking neurons with multiplesynapticconnections，and proposed abiologicallplausibleonlinesupervisedlearning algorithmtosimultaneouslyoptimize thesynapticweightsand synaptic delaysof spiking neurons.Thealgorithmconstructedareal-time error functionusingthekernelfunctionrepresentationofspiketrains，andderivedreal-time updaterulesforsynaptic weightsand synapticdelaysusingthegradientdescent method.Theresultsof spiketrainlearning and nonlinear paternclasificationtasks showthatthedynamicsynapticdelaylearning algorithmhashigherlearningaccuracyandrequires fewerlearming epochs than the static synapticdelaylearning algorithm，andthe learning acuracyof multiple synapticconnectionsis higher thanthatof singlesynapticconnections.Itcanbeseenthatthe synapticdelayplasticityand multiple synapticconnection modecaeffectively improve the learning performance of spiking neural networks.

Key words：spiking neural network（SNN）；online supervised learning;synaptic delay learning;multiple synaptic connections

0 引言

脉冲神经网络（SNN）是一种生物神经系统启发的新一代人工神经网络模型，具有较强的生物可解释性。（剩余17604字）