| 本文已被：浏览 4146次 下载 5758次 |
码上扫一扫！ |
|
|
| 神经网络学习中“附加样本”的技术 |
|
张 铃1,2, 张 钹3,2
|
|
1.安徽大学人工智能研究所,合肥,230039;2.清华大学智能技术与系统国家重点实验室,北京,100084;3.清华大学计算机科学与技术系,北京,100084
|
|
| 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. |
| 关键词: 神经网络,FP算法,附加样本,最小覆盖. |
| DOI： |
| 分类号: |
| 基金项目:本文研究得到国家自然科学基金、国家863高科技项目基金和国家“攀登计划”基金资助. |
|
| A New Learning Algorithm Based on Technique Appended Samples |
|
ZHANG Ling,ZHANG Bo
|
| Abstract: |
| 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(n4)). 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. |
| Key words: Neural network, FP algorithm, technique of appended samples, minimum covering. |
