基于条件互信息的动态贝叶斯法探明生物神经元网络连接结构  

Identification of biological neuron network connection structures by dynamic Bayesian network method based on conditional mutual information

作　　者：任婧雯 董朝轶[1,2] 朱美佳 白鹏辉 赵肖懿 马爽 贾婷婷 REN Jingwen;DONG Chaoyi;ZHU Meijia;BAI Penghui;ZHAO Xiaoyi;MA Shuang;JIA Tingting(School of Electric Power,Inner Mongolia University of Technology,Hohhot 010080,China;Inner Mongolia Key Laboratory of Electromechanical Control,Hohhot 010051,China)

机构地区：[1]内蒙古工业大学电力学院,内蒙古呼和浩特010080 [2]内蒙古机电控制重点实验室,内蒙古呼和浩特010051

出　　处：《中国医学物理学杂志》2021年第6期773-779,共7页Chinese Journal of Medical Physics

基　　金：国家自然科学基金(61863029,61364018);内蒙古自然科学基金杰出青年培育基金(2016JQ07);内蒙古自治区高等学校“青年科技英才计划”(NJYT-15-A05)。

摘　　要：准确辨识生物神经元网络(BNN)连接结构对于进一步理解其网络行为与功能,构建具有生物真实性、结构更加优化的人工智能网络具有重要意义。本文提出基于条件互信息的动态贝叶斯网络法,以探明BNN的连接结构。首先,利用积分点火原理构造脉冲神经元网络,经过网络仿真后得到多通道动态响应数据;然后,针对该数据集,计算神经元节点间的条件互信息,通过与对给定阈值δ进行比较,判断节点间的连接情况;最后,辨识出动态贝叶斯网络的连接结构。实验结果表明基于条件互信息的动态贝叶斯网络法对于BNN具有较高的辨识正确率。The accurate identification of biological neural network(BNN)connection structures helps to further understand their network behaviors and functions,and contributes to constructing biologically realistic artificial intelligent networks with more optimized structures.Dynamic Bayesian network method based on conditional mutual information is proposed for accurately identifying the connection structures of BNN.Spike neural network is firstly constructed by integrate-and-fire principle,and the multi-channel dynamic response data are obtained after network simulation.Based on the obtained data set,the conditional mutual information between neuron nodes are calculated,and the connection between nodes is assessed by comparing the calculated results with the given threshold δ.Finally,dynamic Bayesian network connection structures are identified.The experimental results reveal that dynamic Bayesian network method based on conditional mutual information has a high identification accuracy for BNN.

关 键 词：生物神经网络 贝叶斯网络 条件互信息 积分点火模型 

分 类 号：R318[医药卫生—生物医学工程] Q612[医药卫生—基础医学]