两维局域互联神经网络的关联存储

Two-dimensional local interconnection neural network for associative memory

本文将局域互联神经网络的新概念推广到两维情形,并对两维局域互联关联存储进行了理论分析和大量的计算机模拟.结果表明,两维局域互联神经网络的优点是,在满足存储容量限制的前提下,它与全局互联神经网络具有相同的关联存储能力,而其互联权重矩阵要比全局互联网络小得多.因而,有利于使用现有的空间光调制器实现两维大规模的人工神经网络.

The new concept of local interconnection neural network (LINN) is extended to two-dimensional case. Theoretical analysis and large number of computer simulations on 2-D LINN associative memory are presented. It is concluded that under the limitation of storage capacity LINN has the same associative memory ability as global interconnection neural network (GINN). Moreover, the LINN has a much smaller interconnection weight matrix compared with GINN. Therefore, 2-D LINN makes it possible to realize a large scale neural network by using presently available spatial light modulators.