论文部分内容阅读
为分析人工神经网络（ＡＮＮ）的品质或性能对权值扰动及输入误差的容忍特性，文章基于输入及权值的随机模型，采用统计学的方法，得到了由ｓｉｇｍｏｉｄ型神经元构成的任意多层前馈神经网络在任意大小且具有任意相关性的输入误差与／或权值扰动下，输出误差特性的通用算法。仿真及对比理论计算结果表明了所提算法是正确的。
In order to analyze the quality of performance or performance of artificial neural network (ANN) and its tolerance to the disturbance of weight and input error, a random number model based on input and weight is adopted, and statistical methods are used to obtain any number of sigmoid-type neurons A generalized algorithm for output error characteristics of layer feedforward neural networks with input errors and / or weight perturbations of arbitrary size and with any correlation. Simulation and comparison theory results show that the proposed algorithm is correct.