一种基于大脑皮层结构的侧抑制神经网络  

A lateral inhibition neural network based on neocortex topology

作　　者：杨刚[1] 乔俊飞[1] 薄迎春[1,2] 韩红桂[1] 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100124 [2]中国石油大学信息与控制工程学院,山东青岛266555

出　　处：《控制与决策》2013年第11期1702-1706,共5页Control and Decision

基　　金：国家自然科学基金重点项目(61034008);教育部博士点基金项目(20080005004)

摘　　要：借鉴仿生学原理,基于大脑皮层结构提出一种新型侧抑制神经网络(S-LINN)模型.通过模拟大脑皮层内锥体神经元和抑制神经元的连接特点,在多层结构的S-LINN的不同层神经元之间引入跨越连接,同时在隐含层内神经元之间进行信息的侧向抑制传输.引入的两种连接机制有效地提高了网络处理问题的能力,与其他网络相比能够以更精简的结构较好地解决实际问题.通过对乳腺癌诊断数据集和异或问题的求解,表明了S-LINN网络不但能够获得较高的训练精度,而且具有更强的泛化能力.On the basis of the connective mode and information transmit mechanism of cerebral cortex, a novel lateral inhibition neural network model with span connection（S-LINN） is proposed. Combining the liminar organization of cerebral cortex, and fully considering the lateral inhibition connection between interneurons and the pyramidal neurons, the proposed S-LINN transforms information to other neurons in different layers, which is used to enhance response contrast and advance the network representation, respectively. The effectiveness and superiority of the proposed network is compared with other popular approaches on two benchmark problems in the areas of real-world regression and classification problems. Simulation results show that the proposed S-LINN can achieve the higher accuracy of approximation and generalization with the comparably compact network structure.

关 键 词：仿生学 大脑皮层 侧抑制 人工神经网络 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]