Font Size: a A A

Research And Application Of The Learning Algorithm For Spiking Neural Networks

Posted on:2020-02-05Degree:MasterType:Thesis
Country:ChinaCandidate:Q T XueFull Text:PDF
GTID:2428330596976537Subject:Engineering
Abstract/Summary:PDF Full Text Request
The emergence of Spiking Neural Networks(SNN)narrows the gap between neuroscience and machine learning.SNN uses neural structures which have more biological characteristics,and encodes information with time for carrying more information.Although SNN has achieved some remarkable results in practical applications such as image,vision and speech,SNN is still a relatively new research field at present.Therefore,the research of spiking neural networks has great research value in both theory and application.Research on learning algorithms of spiking neural networks has always been a hot and difficult point in this field.At present,the lack of efficient and accurate learning algorithms greatly limits the practical applications of spiking neural networks,and has become a major bottleneck in spiking neural network research.In order to improve the learning efficiency and accuracy of the spiking neural network,this thesis makes the following contributions:1)This thesis studies the basic theoretical knowledge of spiking neural networks,introduces the common spiking neuron models,and describes the information coding methods of spiking neurons.In addition,this thesis also explain the principle of how spiking neural network identifies timing information.2)This thesis proposes a multi-spike learning algorithm,namely MSLMGD.Although the traditional spike-driven learning algorithms are biological plausible,their learning accuracy and efficiency are relatively low.In order to solve this problem,this thesis proposes a new multi-spike based learning algorithm called MSLMGD.The proposed MSLMGD algorithm uses the membrane potential as the signal to adjust the synaptic weights directly.Experimental results demonstrate that the proposed algorithm has better accuracy and efficiency.3)In order to further improve the performance of the proposed MSLMGD learning algorithm,this thesis proposes a DL-MSLMGD supervised learning algorithm that takes into account synaptic delay plasticity.The traditional learning algorithms of spiking neural networks only consider the plasticity of synaptic weights,and ignore the role of synaptic delay in learning,resulting in low learning performance.Therefore,DLMSLMGD merges the weight update rule and synaptic delay learning rule to enhance the learning performance.4)Finally,the performance of the proposed algorithm is evaluated by comparing with the mainstream algorithm,and the effectiveness of the proposed algorithm is verified by simulation experiments and practical applications.The experimental results show that the proposed algorithm has higher accuracy and efficiency and has good application performance.
Keywords/Search Tags:Spiking Neural Networks, Multi-Spike, Delay, DL-MSLMGD, DLTempotron
PDF Full Text Request
Related items