自动生成多层神经网络的算法及其在编码中的应用

The Algorithm of Automatically Generating Multi-Level Neural Network and Its Application in Encodeing

本文提出了一种多层神经网络自动生成的新算法。对给定的问题,它能够在学习中动态生成一个多层网络,并确定网络的连接方式。通过采用模拟退火的方法,加快了学习的速度。该算法能随时按要求生成神经网络编、译码器,便于处理等长码和变长码问题,还能进行误码纠正,适合于保密通信的需要。

This paper presents a new algorithm of automatically generating multi-level neural network. For a given problem, this algorithm can dynamically generate a multi-level neural network by study and determines the network connecting pattern. In order to accelerate learning, a simulated annealing is used. The algorithm which can generate neural network encoder and decoder, easily processe the problems of the equal length code or the uaequaI length code and correct the error code is therefore very suitable for the requirement of secret communication.