多类别学习方法：理论比较与影响

May, 2012

Multiclass Learning Approaches: A Theoretical Comparison with Implications

Amit Daniely, Sivan Sabato, Shai Shalev Shwartz

TL;DR本论文在五种流行的多分类方法中进行理论分析和比较，并使用 VC 理论分析了这些方法的估计误差和逼近误差，得出了一些实用的结论。

Abstract

We theoretically analyze and compare the following five popular multiclass classification methods: one vs. all, All Pairs, tree-based classifiers