带有增强和变换因子的三层结构双向联想记忆神经网络

提出了一种带有增强和变换因子的三层结构的双向联想记忆神经网络。介绍了增强和变换因子的巧妙选取方法,此方法可使网络很好地适用于相关度很高的样本。但是网络容量的增加往往伴随着寄生稳态点的增加,为避免这种情况,在网络中引入三层结构,这种新的双向联想记忆神经网络能够确保任意相关的样本达到完美的双向联想记忆。仿真实验证明了该方法的优越性。

Three layers structure bidirectional associative memory with dilation and translation parameters

A three layers neural network of bidirectional associated memory (BAM) with dilation and translator parameters is presented. At first a strategy with a clever choice method of dilation and translation parameters is introduced so that the network is better suitable to the highly correlated samples. However the increase of network capasity is accompanied often by an increase of spuriom states. To avoid this situation we introduce a three layers structure into the network. This new BAM neural network can guarantee arbitrary related sample to obtain perfect bidirectional recall. Simulation test has proved its superiority.