ALIM: Adjusting Label Importance Mechanism for Noisy Partial Label Learning

TL;DR

This paper introduces ALIM, a novel framework for noisy partial label learning that mitigates detection errors' impact, achieving state-of-the-art results by adjusting label importance during training.

Contribution

The paper proposes ALIM, a plug-in mechanism with theoretical guarantees, to improve noisy PLL by balancing candidate sets and model outputs, reducing detection error effects.

Findings

Achieves state-of-the-art performance on benchmark datasets.

Effectively reduces negative impact of detection errors.

Demonstrates robustness across various noisy PLL scenarios.

Abstract

Noisy partial label learning (noisy PLL) is an important branch of weakly supervised learning. Unlike PLL where the ground-truth label must conceal in the candidate label set, noisy PLL relaxes this constraint and allows the ground-truth label may not be in the candidate label set. To address this challenging problem, most of the existing works attempt to detect noisy samples and estimate the ground-truth label for each noisy sample. However, detection errors are unavoidable. These errors can accumulate during training and continuously affect model optimization. To this end, we propose a novel framework for noisy PLL with theoretical guarantees, called ``Adjusting Label Importance Mechanism (ALIM)''. It aims to reduce the negative impact of detection errors by trading off the initial candidate set and model outputs. ALIM is a plug-in strategy that can be integrated with existing PLL approaches. Experimental results on benchmark datasets demonstrate that our method can achieve state-of-the-art performance on noisy PLL. \textcolor[rgb]{0.93,0.0,0.47}{Our code can be found in Supplementary Material}.