Combating Noisy Labels through Fostering Self- and Neighbor-Consistency

TL;DR

This paper introduces Jo-SNC, a robust training method for deep learning that effectively identifies and handles noisy labels by leveraging self- and neighbor-consistency, improving accuracy in noisy data scenarios.

Contribution

The paper proposes a novel noise-robust framework combining sample selection, adaptive thresholding, and triplet consistency regularization to better manage label noise and out-of-distribution data.

Findings

Outperforms state-of-the-art methods on benchmark datasets.

Effectively distinguishes clean, noisy in-distribution, and out-of-distribution samples.

Enhances model robustness through triplet consistency regularization.

Abstract

Label noise is pervasive in various real-world scenarios, posing challenges in supervised deep learning. Deep networks are vulnerable to such label-corrupted samples due to the memorization effect. One major stream of previous methods concentrates on identifying clean data for training. However, these methods often neglect imbalances in label noise across different mini-batches and devote insufficient attention to out-of-distribution noisy data. To this end, we propose a noise-robust method named Jo-SNC (\textbf{Jo}int sample selection and model regularization based on \textbf{S}elf- and \textbf{N}eighbor-\textbf{C}onsistency). Specifically, we propose to employ the Jensen-Shannon divergence to measure the ``likelihood'' of a sample being clean or out-of-distribution. This process factors in the nearest neighbors of each sample to reinforce the reliability of clean sample identification. We design a self-adaptive, data-driven thresholding scheme to adjust per-class selection thresholds. While clean samples undergo conventional training, detected in-distribution and out-of-distribution noisy samples are trained following partial label learning and negative learning, respectively. Finally, we advance the model performance further by proposing a triplet consistency regularization that promotes self-prediction consistency, neighbor-prediction consistency, and feature consistency. Extensive experiments on various benchmark datasets and comprehensive ablation studies demonstrate the effectiveness and superiority of our approach over existing state-of-the-art methods.