Neuromorphic Implementation of Visual Cognitive Functions in A Two-variable Winner-take-all Circuit with NMDARs

详细信息    查看官网全文

• 英文论文题名：Neuromorphic Implementation of Visual Cognitive Functions in A Two-variable Winner-take-all Circuit with NMDARs
• 论文作者：游宏志
• 英文论文作者：Hongzhi You ; Key Laboratory for Neuro Information of Ministry of Education ; Center for Information in Bio Medicine ; School of Life Science and Technology ; University of Electronic Science and Technology of China
• 年：2016
• 作者机构：Key Laboratory for Neuro Information of Ministry of Education, Center for Information in Bio Medicine, School of Life Science and Technology, University of Electronic Science and Technology of China;
• 英文论文关键词：Winner-take-all competition ; Neuromorphic engineering ; Visual cognitive functions
• 会议召开时间：2016-08-04
• 会议录名称：第三届全国神经动力学学术会议论文摘要集
• 语种：英文
• 分类号：Q42
• 学会代码：AGLU
• 会议名称：第三届全国神经动力学学术会议
• 会议地点：中国甘肃敦煌
• 主办单位：中国力学学会动力学与控制专业委员会神经动力学专业组
• 学会名称：中国力学学会
• 页数：1
• 文件大小：611k
• 原文格式：D
• 会议级别：全国

摘要

Winner-take-all(WTA) competition is an important computational principle in the brain, by which neurons can compete with each other for activation. Through configuring its network parameters, the WTA neural network can achieve various dynamic characteristics of attractors underlying many visual cognitive functions, such as decision making, working memory and hysteresis in visual perceptions. Therefore, physical implementations of the WTA network with the N-methyl-D-aspartate receptors(NMDARs) in silicon is of benefit to design elementary neuromorphic modules of these visual cognitive functions. In our work, we developed a WTA circuit with NMDARs with desired dynamics through the dynamical system approach of circuit synthesis according to the two-variable version of a biophysical plausible WTA model. Through comparing results from circuit simulations in Cadence and theoretical analysis in dynamical systems approach, we demonstrated that the WTA circuit we build could(1) implement decision task and reproduce ramping neural activities observed in the electrophysiological recording in monkey's experiments as attractor dynamics of the WTA model predicts,(2) reproduce the sustained activities and the limited memory capacity during working memory tasks,(3) achieve hysteresis observed in the visual perception and operatefollowing the theoretical analysis for the WTA model. Neuromorphic implementation of these visual cognitive functions in the WTA circuit will be benefit to develop new generation of technologies that carry out brain-like artificial intelligence while maintaining remarkable energy efficiency.
Winner-take-all(WTA) competition is an important computational principle in the brain, by which neurons can compete with each other for activation. Through configuring its network parameters, the WTA neural network can achieve various dynamic characteristics of attractors underlying many visual cognitive functions, such as decision making, working memory and hysteresis in visual perceptions. Therefore, physical implementations of the WTA network with the N-methyl-D-aspartate receptors(NMDARs) in silicon is of benefit to design elementary neuromorphic modules of these visual cognitive functions. In our work, we developed a WTA circuit with NMDARs with desired dynamics through the dynamical system approach of circuit synthesis according to the two-variable version of a biophysical plausible WTA model. Through comparing results from circuit simulations in Cadence and theoretical analysis in dynamical systems approach, we demonstrated that the WTA circuit we build could(1) implement decision task and reproduce ramping neural activities observed in the electrophysiological recording in monkey's experiments as attractor dynamics of the WTA model predicts,(2) reproduce the sustained activities and the limited memory capacity during working memory tasks,(3) achieve hysteresis observed in the visual perception and operatefollowing the theoretical analysis for the WTA model. Neuromorphic implementation of these visual cognitive functions in the WTA circuit will be benefit to develop new generation of technologies that carry out brain-like artificial intelligence while maintaining remarkable energy efficiency.

引文

[1]Wong,K.-F.and Wang,X.-J.,A recurrent network mechanism of time integration in perceptualdecisions,The Journal of neuroscience,2006,26,4,1314–1328.
    [2]You,H.and Wang,D.,Neuromorphic implementation of attractor dynamics in decision circuit with nmdars,Circuits and Systems(ISCAS),2016 IEEE International Symposium on,2016,2708–2711.
    [3]Arthur,J.V.and Boahen,K.,Silicon-neuron design:A dynamical systems approach,Circuits and Systems I:Regular Papers,IEEE Transactions on,2011,58,5,1034–1043.