摘要
首先详细分析了固定型连接故障和错误输入信号对单个神经元的影响，然后研究了具有硬限幅作用函数的前向神经网络在存在固定型故障情况下的性能变化，并给出了性能与连接故障数之间的关系式及故障性能的分析方法。经用ＭｏｎｔｅＣａｒｌｏ方法仿真表明，文中提出的方法是正确的。最后还对分析的结果进行了讨论，说明这类网络的规模越小，对提高其容错性能越有利。
In this paper, firstly,the effect of stuck at faults and error inputs on aingle neural network node is analysed in detail. Next, the behavior of feedforward neural networkswith hard-limit activation function under stuck-at faults is studied, and the relationship between the network performance and the number of faulty interconnections, together with themethod for analyzing the faulty behavior, is presented. The Monte Carlo simulations showthat the method proposed in this paper is correct. Lastly, the analysis results indicate thatthe smaller the scale of such networks is, the more fault--tolerant they will be.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
1995年第4期88-94,共7页
Journal of Tsinghua University(Science and Technology)
基金
国家自然科学基金
关键词
前向神经网络
故障性能
容错性
神经网络
feedforward neural networks
faulty behavior
fault tolerance
central Limit theorem
