| 基于角度的分类方法综述 |
| |
| 作者姓名： | 付盛 薛原 张三国 |
| |
| 作者单位： | 中国科学院大学数学科学学院, 北京 100049 |
| |
| 基金项目： | Supported by the Special Fund of University of Chinese Academy of Sciences for Scientific Research Cooperation (Y652022Y00) |
| |
| 摘 要： | 统计分类问题经常出现在很多应用中，如人脸识别、欺诈检测和手写字符识别等。对有监督分类问题的统计方法进行综述。特别地，介绍基于角度的分类结构，将二分类与多分类问题纳入一个统一的框架中。讨论基于角度分类器的若干新变体，如稳健学习和加权学习。此外，还指出这些分类器关于Fisher相合性的若干理论结果。
|
| 关 键 词： | 基于角度的分类框架 Fisher相合性 稳健学习 统计分类 加权学习 |
| 收稿时间： | 2018-01-02 |
| 修稿时间： | 2018-01-02 |
Survey on angle-based classification |
| |
| Authors: | FU Sheng XUE Yuan ZHANG Sanguo |
| |
| Institution: | School of Mathematical Sciences, University of Chinese Academy of Sciences, Beijing 100049, China |
| |
| Abstract: | Statistical classification problems are widely encountered in many applications, e.g., face recognition, fraud detection, and hand-written character recognition. In this article we make a comprehensive analysis on statistical methods for supervised classification problems. Specifically, we introduce the angle-based classification structure, which combines binary and multicategory problems in a unified framework. Several new variants of the angle-based classifiers are also discussed, such as robust learning and weighted learning. Furthermore, we show some theoretical results about Fisher consistency for these angle-based classifiers. |
| |
| Keywords: | angle-based classification framework Fisher consistency robust learning statistical classification weighted learning |
|
| 点击此处可从《》浏览原始摘要信息 |
| 点击此处可从《》下载免费的PDF全文 |
