两种典型神经网络容错方法的比较

Behnam提出的SC算法和文中提出的rehidden算法是两种典型的前向神经网络容错 算法,前者改进BP算法进行学习,后者对已学习的网络进行隐层节点冗余.这两种算法各有优 缺点.文中对这两种算法进行了仿真实验分析,最终得到了每种算法适用的网络规模和硬件条 件,在不同环境下应采用不同的方法才能得到可行的容错网络.最后还对SC算法的一些改进进 行了讨论.

COMPARISON OF TWO TYPICAL FAULT-TOLERANCE ALGORITHMS OF NEURAL NETWORKS

There are two typical fault tolerance algorithms of feed forward neural networks. One is SC algorithm presented by Behnam and the other is rehidden algorithm presented by this paper. The former modifies the BP algorithm in the training phase to gain fault tolerant network, and the latter gives some redundant nodes to the hidden layer of the trained neural network. There are advantage and shortage in both algorithms. We do simulations on these two algorithms. Analysis of the simulation results show that either algorithm has its own applicable network scale and hardware condition. In different condition, different algorithm should be used to gain suitable fault tolerant neural network. Finally, we also give some analysis to some improvement of SC algorithm.