XMix: Combating Extremely Noisy Labels via Local Smoothness in Self-Supervised Feature Space

2026-07-26

AI Summary

This work addresses the significant performance degradation of supervised learning under extremely noisy labels, severe class imbalance, and the absence of noise priors. To tackle this challenge, the authors propose XMix, a novel framework that uniquely integrates the local smoothness of self-supervised feature spaces into the entire noisy-label learning pipeline. Without relying on corrupted labels or prior knowledge of noise rates, XMix robustly estimates noise rates via neighborhood-based maximum likelihood, performs balanced sample selection with neighborhood assistance, and generates pseudo-labels driven by neighborhood consistency. Extensive experiments demonstrate that XMix substantially outperforms existing methods on both extreme noise settings and standard learning-with-noisy-labels benchmarks, exhibiting superior robustness and generalization capability.

Abstract

Supervised deep learning models rely on large, accurately labeled datasets, yet noisy annotations are often unavoidable and can severely degrade performance under high noise levels. Recent state-of-the-art methods tackle this by using sample selection strategies that exploit the memorization effect to filter out clean data for semi-supervised learning. However, these methods struggle with extreme noise, class imbalance, and require careful tuning or prior noise knowledge. To address these limitations, we propose XMix, a novel framework that leverages local smoothness in the self-supervised feature space to systematically enhance all stages of the sample selection process, without dependence on potentially corrupted labels. First, XMix estimates the noise rate using maximum likelihood among self-supervised feature neighbors. Second, these neighbors then help identify additional clean samples and ensure balanced selection across classes during sample selection. Finally, in the semi-supervised learning phase, XMix uses neighboring samples to generate more reliable pseudo-labels. Our empirical results show that XMix substantially outperforms existing methods in extremely noisy environments and maintains superior performance in standard LNL benchmarks.

Problem

• noisy labels
• extreme noise
• class imbalance
• sample selection
• supervised learning

Innovation

• local smoothness
• self-supervised feature space
• noisy label learning
• sample selection
• pseudo-labeling