论文部分内容阅读
本文基于Hopfield神经网络模型,实现了可编程光电混合的二维二次关联记忆。在实验系统中,分别以液晶电光开关阵列及液晶电视实现实时输入及权重矩阵显示,并以体全息透镜阵列实现三维空间互连。本文提出并采用了一种分时处理方法,简化了系统并降低了对空间光调制器件的容量要求,为实现更多神经元的关联记忆提供了一条可能的途径。文中给出了初步实验结果。
Based on the Hopfield neural network model, this paper realizes two-dimensional quadratic associative memory of programmable optoelectronic hybrid. In the experimental system, real-time input and weight matrix display are respectively realized by the liquid crystal electro-optical switch array and the liquid crystal television, and the three-dimensional space interconnection is realized by the volume holographic lens array. This paper proposes and adopts a time-sharing approach, which simplifies the system and reduces the capacity requirements for spatial light modulator devices. This provides a possible way to realize the associated memory of more neurons. The preliminary experimental results are given in this paper.