不对称神经网络的自旋玻璃模型和相变温度

A Spin-glass Model of Asymmetrical Between Pre-and Post-Synapse Neural Network and Transition Temperature

本文将文献[1]的突触前后不对称神经网络推广到有随机耦合的自旋玻璃模型中去.证明了引入突触前因子可以提高相变温度,其机理在于此因子分布的弥散性.特别是对于单向传递的多层网络,由于相邻层突触前因子有很大差别,而使得相变温度可有效提高.

The neural neetwork model of asymmetrical between pre-and post-synapae propesed in ref.1 is generalized to the case of random coupling.Acorresponding spin-glass model is studied in detail.It is proved that the introduction of pre-synapse factors and the large variance of their dostributopn would cause an increase of the transition temperature of the network.For one-directional multi-layered net- work the transition temperature can be raised in a large amount due to the large difference of the pre- synapse factors in neighboring layers.