A theory of multiclass boosting

TL;DR

This paper develops a comprehensive theoretical framework for multiclass boosting, identifying optimal weak classifier requirements and designing the most effective boosting algorithms for multiclass problems.

Contribution

It introduces a broad, general framework for multiclass boosting, specifying weak classifier requirements and designing optimal algorithms, filling a gap in theoretical understanding.

Findings

Defines optimal weak classifier requirements for multiclass boosting

Designs the most effective boosting algorithms under these requirements

Provides a theoretical foundation for future multiclass boosting methods

Abstract

Boosting combines weak classifiers to form highly accurate predictors. Although the case of binary classification is well understood, in the multiclass setting, the "correct" requirements on the weak classifier, or the notion of the most efficient boosting algorithms are missing. In this paper, we create a broad and general framework, within which we make precise and identify the optimal requirements on the weak-classifier, as well as design the most effective, in a certain sense, boosting algorithms that assume such requirements.