###
DOI:
Journal of Software:1993.4(4):56-60

一个基于信息论的示例学习方法
钟鸣,陈文伟,张凯慈
(解放军防化研究院计算中心,北京 102205;国防科学技术大学,长沙 410073;解放军防化研究院计算中心 北京 102205)
AN INFORMATION—BASED METHOD IBLE FOR LEARNING FROM EXAMPLES
Zhong Ming;,Chen Wenwei,Zhang Kaici
()
Abstract
Chart / table
Reference
Similar Articles
Article :Browse 2979   Download 2945
Received:May 01, 1991    
> 中文摘要: 本文利用信息论中信道容量、最大似然译码准则等概念,提出一个新的示例学习方法IBLE,此方法不依赖类别先验概率,特征间为强相关,具有直观的知识表示,将它用于质谱解析,结果很好,八类化合物平均正确预测率为93.96%,高于专家水平。
中文关键词:
Abstract:This paper presents a new method IBLE for learning from examples with the concepts of capacity, maximal plausible decode criterion of information theory. The method doesn t depend on the prior probability of class. In the method,the attributes are strongly associated, the knowledge representation is intelligible. We use IBLE in the interpretation of mass spectra, good result is obtained and the average predictive accuracy for eight classes of compounds is 93. 96%. This result is superior to experts.
keywords:
文章编号:     中图分类号:    文献标志码:
基金项目:
Foundation items:
Reference text:

钟鸣,陈文伟,张凯慈.一个基于信息论的示例学习方法.软件学报,1993,4(4):56-60

Zhong Ming;,Chen Wenwei,Zhang Kaici.AN INFORMATION—BASED METHOD IBLE FOR LEARNING FROM EXAMPLES.Journal of Software,1993,4(4):56-60