【摘要】 利用神经网络提取含噪数据的特征时，如何提高泛化能力是亟需解决的重要问题。通常的解决方法是通过结构优化和正则化来控制网络的复杂性。本文从熵函数的概念出发，提出了一项新的加性惩罚因子以消除过训练现象，从而有效的提高网络的泛化能力。更多还原

【Abstract】 Generalization ability is a major problem encountered when using neural network to find the structures in a noisy data sets.Controlling the network complexity is a commom method to solve the problem.In this paper,however,a novel additive penalty term which represents the features extracted by hidden units is introduced to eliminate the overtraining of multiayer feedfoward networks.Computer simulations demonstrate that the generalization ability is greatly improved.更多还原