论文部分内容阅读
本文对用于模式分类、函数逼近、参数估计的多层感知器 (MLPs)给出 1个清晰的关于内部行为的解释。作者以单隐层的 MLP为例 ,论述了关于 MLP的内部行为的半线性分析理论。对受训的MLP,将隐层单元的输出分别定义为网络输出的正、负“内部分量”;定义内部分量的连接权重集为给定问题的“内部判别模式”;建立了 MLP和模糊集相结合的新模型 ;分析了 MLP的结构为 N- 2 - 1和N- H- 1 ,给出权重初始化的方法 ;提出了 1种从受训神经 -模糊模型 (NFMs)中提取知识的全新的具有实用价值的方法。
This paper gives a clear explanation of the internal behavior of MLPs for pattern classification, function approximation and parameter estimation. Taking a single hidden layer MLP as an example, the author discusses the semi-linear analysis theory about the internal behavior of MLP. For the trained MLP, the output of hidden layer units are respectively defined as the positive and negative “internal components” of the network output; the “internal discriminant model” for defining the connection weights of internal components as the given problem; and the MLP and fuzzy sets A new model of MLP is proposed. The structure of MLP is analyzed as N - 2 - 1 and N - H - 1, and a weight initialization method is given. A new method of extracting knowledge from trainee neuro - fuzzy models (NFMs) Practical value of the method.