ChronoSelect: Robust Learning with Noisy Labels via Dynamics Temporal Memory

TL;DR

ChronoSelect introduces a temporal memory framework that dynamically leverages learning evolution to improve robustness against noisy labels in deep neural networks.

Contribution

It proposes a novel four-stage memory architecture with a sliding update mechanism to better distinguish clean, boundary, and noisy samples during training.

Findings

Achieves state-of-the-art results on synthetic benchmarks.

Effectively partitions data into clean, boundary, and noisy subsets.

Demonstrates theoretical convergence guarantees.

Abstract

Training deep neural networks on real-world datasets is often hampered by the presence of noisy labels, which can be memorized by over-parameterized models, leading to significant degradation in generalization performance. While existing methods for learning with noisy labels (LNL) have made considerable progress, they fundamentally suffer from static snapshot evaluations and fail to leverage the rich temporal dynamics of learning evolution. In this paper, we propose ChronoSelect (chrono denoting its temporal nature), a novel framework featuring an innovative four-stage memory architecture that compresses prediction history into compact temporal distributions. Our unique sliding update mechanism with controlled decay maintains only four dynamic memory units per sample, progressively emphasizing recent patterns while retaining essential historical knowledge. This enables precise three-way sample partitioning into clean, boundary, and noisy subsets through temporal trajectory analysis and dual-branch consistency. Theoretical guarantees prove the mechanism's convergence and stability under noisy conditions. Extensive experiments demonstrate ChronoSelect's state-of-the-art performance across synthetic and real-world benchmarks.