关键词: 神经网络 神经元 多层感知机 超平面 分类

Abstract:

Multilayer perceptrons are widely applied in classification. This paper concentrates on the hyperplane multilayer perceptrons for classification. And the maximum number of space cells produced by input neurons is formulated. It shows the classification ability of the input neurons. Based on this, a mo re accurate upper limit of the hidden layer neuron number is achieved by discussing the restriction between the input neuron number and the hidden neuro n number. It is proved that the upper limit of the hidden layer neuron number in this paper is much smaller than the former results when the input space dimension is far smaller than the input layer neuron number.

Key words: （neural network, neurom multilayer perceptrom hyperplane, classification）