| 摘要: |
| 在讨论神经网络的容错性的文献中,主要涉及的一直是关于输入噪音的容错问题.在这些文献中通常把该问题转换为某种优化问题,并用现成的优化方法进行求解,但很少涉及由网络故障所引起的容错问题,即结构容错问题.利用覆盖算法分析结构容错问题,给出一个神经网络容纳所有单节点故障的充要条件和构造这种网络的算法.这些结果揭示了神经网络结构容错能力的本质,并提供了一种分析神经网络容错的新方法. |
| 关键词: 神经网络 容错性 结构容错性 单点故障 |
| DOI： |
| 分类号: |
| 基金项目:国家自然科学基金资助项目(69675011);国家重点基础研究发展规划973资助项目(G1998030509) |
|
| On the Fault Tolerance Problem of the Feedforward Neural Networks |
|
ZHANG Ling
|
| Abstract: |
| Until recently, in the literatures of discussing the fault tolerance problem of neural networks, the tolerance about input noise is mainly concerned. The problem is generally transformed into that of optimization and solved by some well-known optimization approach. But only a few dealt with the tolerance due to the network structural failure, i.e., the structural fault tolerance. In the paper, the structural fault tolerance is analyzed by using the covering algorithms. The necessary and sufficient conditions of allowing of single node failure in a neural network and the algorithm for constructing such a network are given. These results reveal the essence of the structural fault tolerance capacity and show a new way for the analysis of the fault tolerance of neural networks. |
| Key words: neural network fault tolerance structural fault tolerance single fault |
