Learning to Clean: Reinforcement Learning for Noisy Label Correction

TL;DR

This paper presents RLNLC, a reinforcement learning framework that automatically corrects noisy labels in datasets, significantly improving model training accuracy in noisy label scenarios.

Contribution

It introduces a novel RL-based approach for label correction, integrating a deep policy network with an actor-critic method for improved noise handling.

Findings

RLNLC outperforms existing methods on benchmark datasets.

The approach effectively reduces label noise impact.

Demonstrates robustness across multiple datasets.

Abstract

The challenge of learning with noisy labels is significant in machine learning, as it can severely degrade the performance of prediction models if not addressed properly. This paper introduces a novel framework that conceptualizes noisy label correction as a reinforcement learning (RL) problem. The proposed approach, Reinforcement Learning for Noisy Label Correction (RLNLC), defines a comprehensive state space representing data and their associated labels, an action space that indicates possible label corrections, and a reward mechanism that evaluates the efficacy of label corrections. RLNLC learns a deep feature representation based policy network to perform label correction through reinforcement learning, utilizing an actor-critic method. The learned policy is subsequently deployed to iteratively correct noisy training labels and facilitate the training of the prediction model. The effectiveness of RLNLC is demonstrated through extensive experiments on multiple benchmark datasets, where it consistently outperforms existing state-of-the-art techniques for learning with noisy labels.