Disambiguation of weak supervision with exponential convergence rates

TL;DR

This paper introduces an empirical disambiguation algorithm for weak supervision with partial labels, proving exponential convergence rates and demonstrating its effectiveness on practical examples.

Contribution

It presents a novel disambiguation algorithm for weak supervision with partial labels and establishes its exponential convergence under classical learnability assumptions.

Findings

Algorithm achieves exponential convergence rates.

Method effectively disambiguates partial labels in practice.

Provides theoretical guarantees for weak supervision disambiguation.

Abstract

Machine learning approached through supervised learning requires expensive annotation of data. This motivates weakly supervised learning, where data are annotated with incomplete yet discriminative information. In this paper, we focus on partial labelling, an instance of weak supervision where, from a given input, we are given a set of potential targets. We review a disambiguation principle to recover full supervision from weak supervision, and propose an empirical disambiguation algorithm. We prove exponential convergence rates of our algorithm under classical learnability assumptions, and we illustrate the usefulness of our method on practical examples.