全双极WTA神经网络模型的光学实现

本文给出一种用光学方法实现全双极WTA三层神经网络模型的新方案及其实验结果.在全双极情况下.中间层第i个神经元的输入值由两部分组成,一部分为单极神经元状态时中间层该神经元的输入值,另一部分为对应存储模式的反模式和的二分之一.其中常数1/2可通过将每一输入神经元分成相等的两部分,使其中一部分为全透,另一部分为完全不透来实现.实验表明全双极系统比原先的单极系统具有更高的存储容量和寻址能力.

Optical Implementation of WTA Neural Network with Bipolar Neuron States

The optical implementation of WTA neural network model with bipolar input-layer neurons and bipolar input-layer-to-hidden-layer connections is presented. The input to the hidden-layer neurons is the bipolarly weighted summation of the bipolar neuron states of the input-layer, which is equivalent to the unipolarly weighted summation of the unipolar input states and the summation of the unipolar reversed storage pattern by a constant 1/2. The constant 1/2 can be implemented by dividing each neural pixel into two equal parts of which one is opaque and the other is transparent. Experimental results shows that both the storage capacity and the content addressability of the optical system are improved compared to those of the unipolar WTA optical system