FINE Samples for Learning with Noisy Labels

TLDR

Deep neural networks degrade when training data contain noisy labels, and while noise‑cleansing methods are currently the most effective, their detectors often rely on heuristic thresholds rather than principled theory. This work introduces a new detector for filtering noisy labels. The proposed FINE framework analyzes latent representation dynamics via eigendecomposition of the data Gram matrix to detect noisy instances, offering derivative‑free, theoretically grounded methods and enabling sample‑selection, semi‑supervised learning, and integration with noise‑robust loss functions. Experiments demonstrate that FINE consistently outperforms baseline approaches across sample‑selection, semi‑supervised learning, and loss‑function integration on multiple benchmark datasets.

Abstract

Modern deep neural networks (DNNs) become frail when the datasets contain noisy (incorrect) class labels. Robust techniques in the presence of noisy labels can be categorized into two folds: developing noise-robust functions or using noise-cleansing methods by detecting the noisy data. Recently, noise-cleansing methods have been considered as the most competitive noisy-label learning algorithms. Despite their success, their noisy label detectors are often based on heuristics more than a theory, requiring a robust classifier to predict the noisy data with loss values. In this paper, we propose a novel detector for filtering label noise. Unlike most existing methods, we focus on each data's latent representation dynamics and measure the alignment between the latent distribution and each representation using the eigendecomposition of the data gram matrix. Our framework, coined as filtering noisy instances via their eigenvectors (FINE), provides a robust detector with derivative-free simple methods having theoretical guarantees. Under our framework, we propose three applications of the FINE: sample-selection approach, semi-supervised learning approach, and collaboration with noise-robust loss functions. Experimental results show that the proposed methods consistently outperform corresponding baselines for all three applications on various benchmark datasets.