仅用激发神经元的IPA神经网络模型及其光学实现  

OPTICAL IMPLEMENTATION OF IPA NEUBAL NETWORK MODEL WITH EXCITATION NEURONS

作　　者：王文陆[1] 严瑛白[1] 刘树田[2] 吴杰[2] 李淳飞[2] 

机构地区：[1]清华大学精密仪器与机械学系,100084 [2]哈尔滨工业大学应用物理系,150006

出　　处：《应用光学》1993年第1期37-42,共6页Journal of Applied Optics

摘　　要：仅用激发神经元的IPA(Interpattern Association)型神经网络模型与既有激发又有抑制的IPA模型具有相似的性能。仅用激发神经元后,不需要光强的相减,这样可以较简单地实现全光神经网络系统,对相似的存储模式有较强的分辨力。此文提出了一套二维(8×8)光学神经网络实验系统,用透镜阵列实现互连,并给出了理论描述和光学实验结果。The IPA (interpattern association) neural network model with only excitatory neurons is shown to be similar to networks with both inhibitory and excitation interconnects in performance. An all-optical neural network with the excitatory neurons can be realized simply without the substraction of light intensities at the input ports of the neurons. Compared with the Hopfieldtype model, it is of higher recognition ability even for the patterns being similar to each other. Theory description and optical experiment results are presented in this paper.

关 键 词：神经网络 模式识别 IPA模型 

分 类 号：O438[机械工程—光学工程]