Provably Consistent Partial-Label Learning

TL;DR

This paper introduces a theoretical framework for partial-label learning, proposing a generation model for candidate label sets and two consistent learning methods, validated through experiments on benchmark and real-world datasets.

Contribution

It presents the first generation model for candidate label sets and develops two provably consistent PLL methods compatible with deep networks.

Findings

The proposed generation model accurately describes candidate label sets.

The two PLL methods are risk- and classifier-consistent.

Experiments demonstrate the effectiveness of the methods on various datasets.

Abstract

Partial-label learning (PLL) is a multi-class classification problem, where each training example is associated with a set of candidate labels. Even though many practical PLL methods have been proposed in the last two decades, there lacks a theoretical understanding of the consistency of those methods-none of the PLL methods hitherto possesses a generation process of candidate label sets, and then it is still unclear why such a method works on a specific dataset and when it may fail given a different dataset. In this paper, we propose the first generation model of candidate label sets, and develop two novel PLL methods that are guaranteed to be provably consistent, i.e., one is risk-consistent and the other is classifier-consistent. Our methods are advantageous, since they are compatible with any deep network or stochastic optimizer. Furthermore, thanks to the generation model, we would be able to answer the two questions above by testing if the generation model matches given candidate label sets. Experiments on benchmark and real-world datasets validate the effectiveness of the proposed generation model and two PLL methods.