| 多分类贪婪算法的一致性 |
| |
| 引用本文： | 陈洪,蔡佳,张玉成.多分类贪婪算法的一致性[J].湖北大学学报(自然科学版),2005,27(4):317-320,335. |
| |
| 作者姓名： | 陈洪 蔡佳 张玉成 |
| |
| 作者单位： | 湖北大学,数学与计算机科学学院,湖北,武汉,430062;深圳信息职业技术学院,广东,深圳,518029 |
| |
| 基金项目： | 国家自然科学基金（10371033）资助课题 |
| |
| 摘 要： | 学习理论中，许多学习算法可以描述为一个最小化适当损失函数的贪婪过稗．贪婪算法小依赖于所估计问题的参数的数目，在处理较弱条件的统计估计问题中具有较大的优势．本文研究基于凸风险最小化方法的多分类贪婪算法，推广二分类的学习问题到多分类的情形，建立了多分类贪婪算法的估计误差，证明了该学习算法的一致性。
|
| 关 键 词： | 学习理论 贪婪算法 Redmacher复杂度 |
| 文章编号： | 1000-2375（2005）04-0317-04 |
| 收稿时间： | 2004-11-19 |
| 修稿时间： | 2004-11-19 |
The consistency of greedy algorithms for multi-category classification |
| |
| CHEN Hong, CAI Jia , ZHANG Yu-cheng.The consistency of greedy algorithms for multi-category classification[J].Journal of Hubei University(Natural Science Edition),2005,27(4):317-320,335. |
| |
| Authors: | CHEN Hong CAI Jia ZHANG Yu-cheng |
| |
| Institution: | 1. School of Mathematics and Computer Science, Hubei University, Wuhan 430062, China; 2. Shenzhen Institute of Information Technology, Shenzhen 518029, C hina |
| |
| Abstract: | Learning algorithms,in statistical learning theory,can be described as greedy procedures for stagewise minimization of an appropriate cost funcation.Compared with other procedures,greedy algorithms don't depend on the condition number of associated parametric estimation.This advantage is important since many statistical estimation problem are know to be ill-conditioned.In this paper,learning algorithm for solving multi-category classification using convex upper losses is studied.Estimation error for multi-category classification is established.Finally,the consistency of greedy algorithm is proved. |
| |
| Keywords: | learning theory greedy algorithm redemacher complexity |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
|
