模糊线性判别函数与权重初始化超球面

将非线性神经元及多层感知机分类行为分析建筑在模糊集理论基础上，提出模糊线性判别函数、模糊判别面、模糊模式分类等概念，并引导出将多层感知机的隐层权重值均匀地分布在权重空间超球面上的网络初始化方法。以一系列实验验证此方法能明显提高多层感知机收敛性能，且与所用的学习算法、神经元的激励函数形式无关。

Fuzzy Linear Discriminant Functions and the Weight-Initialization Hyperspheres

Based on the fuzzy set theory, the concepts of fuzzy linear discriminant function, fuzzy decision surface, and fuzzy classification are introduced in this paper. Then, a novel weight-initialization method, by which the input-to-hidden weight vectors of MLP-classifiers are initialized and distributed uniformly on a hypersphere in the weight space, is induced. Using a series of experiments, it is proved that this weight-initialization method is an universal approach to improving markedly the performance of MLP-classifiers, no matter what kind of learning algorithms are exploited and what kind of activation functions are built into the hidden-neurons.