一种改进型HOPFIELD神经网络模型及其单通道光学实现

An Improved HOPFIELD Nerve Network Model and the Realization of its Single Channel Optics

提出一种改进型HOPFIELD神经网络模型。通过对存贮模式进行互补扩展,消除了存贮模式中0和1个数不等问题。利用扩展模式互补性和由扩展模式所形成的连接权的镜象对称性,在不增加神经元个数和连接权矩阵维数的情况下,提高了网络的存贮能力和容错能力。在此基础上设计了全正光学连接权矩阵,在单通道内实现了双极寻址,降低了光学系统的复杂性。

An improved type of HOPFIELD nerve network model has been put forword. The problem of unequal amounts of o and 1 in storage mode can be eliminated by a complementary extension of the storage mode. Using the complementarity of extending mode and the mirror symmetry of connection weight formed by the extending mode,the storage capability of the network can be increased without the increase in the number of nerve-element and the dimensionality of connection weight matrix.On this basis,we have designed an all-positive optical connection weight matrix and realized bipolar addressing in a single channel and decreased the complexity of its optical sysiem.