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

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 overcome these limitations, we propose XMix, which exploits local smoothness in the self-supervised feature space to strengthen sample selection. XMix estimates noise rates via maximum likelihood among feature neighbors, expands and balances the clean set using label-consistent neighbors, and generates more reliable pseudo-labels during semi-supervised learning. Our empirical results show that XMix substantially outperforms existing methods in extremely noisy environments and maintains superior performance in standard LNL benchmarks.