神经网络学习中“附加样本”的技术  

A New Learning Algorithm Based on Technique Appended Samples

作　　者：张铃[1,2,3] 张钹[1,2,3] 

机构地区：[1]安徽大学人工智能研究所 [2]清华大学计算机科学与技术系 [3]清华大学智能技术与系统国家重点实验室

出　　处：《软件学报》1998年第5期371-377,共7页Journal of Software

基　　金：国家自然科学基金;国家863高科技项目基金;国家"攀登计划"基金

摘　　要：本文将网络的先验知识和网络的样本集知识有机结合起来，提出“附加样本”的神经网络新学习算法，其计算复杂性仅为多项式（上界≤Ｏ（ｎ４），用该算法可以设计出性能更好的神经网络．本文第１节简单介绍ＦＰ算法以及ＦＰ覆盖算法，第２节提出ＦＰ统计附加样本算法．最后举一例子说明用该算法可以设计出性能良好的网络．A new learning algorithm of neural network, called technique of appended samples, is proposed in this paper, which well combines a priori knowledge with the training samples knowledge to design a neural network. Its complexity is just polynomial (O(n 4)). This new method can be used in designing a better performance of neural network. This article is structured as follows: the FP algorithm and the cover algorithm based on FP algorithm are introduced in the section 1. In the next section, the FP statistical technique of appended samples is provided. Finally, a design example is given to illustrate the efficiency of new method.

关 键 词：神经网络 FP算法 附加样本 算法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]