TailedCore: Few-Shot Sampling for Unsupervised Long-Tail Noisy Anomaly Detection

2025-04-03

AI Summary

This work addresses the realistic challenge in unsupervised anomaly detection where normal samples are simultaneously corrupted by pixel-level noise and exhibit an unknown, long-tailed class distribution. To resolve the inherent trade-off between noise robustness and tail-class representation, we propose a novel decoupling paradigm. Methodologically, we introduce TailSampler—the first category-size predictor leveraging the symmetry assumption of embedding similarity—to enable precise few-shot sampling for tail classes. We further design TailedCore, a memory-augmented model that jointly integrates memory mechanisms and contrastive learning to enhance both tail-class discriminability and pixel-level noise robustness. Evaluated on a comprehensive long-tailed noisy anomaly detection benchmark, our approach surpasses all state-of-the-art methods. It is the first to systematically alleviate the conflict between tail-class recognition accuracy and noise resilience, delivering a scalable and interpretable unsupervised solution for real-world industrial defect detection.

Technology Category

• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

We aim to solve unsupervised anomaly detection in a practical challenging environment where the normal dataset is both contaminated with defective regions and its product class distribution is tailed but unknown. We observe that existing models suffer from tail-versus-noise trade-off where if a model is robust against pixel noise, then its performance deteriorates on tail class samples, and vice versa. To mitigate the issue, we handle the tail class and noise samples independently. To this end, we propose TailSampler, a novel class size predictor that estimates the class cardinality of samples based on a symmetric assumption on the class-wise distribution of embedding similarities. TailSampler can be utilized to sample the tail class samples exclusively, allowing to handle them separately. Based on these facets, we build a memory-based anomaly detection model TailedCore, whose memory both well captures tail class information and is noise-robust. We extensively validate the effectiveness of TailedCore on the unsupervised long-tail noisy anomaly detection setting, and show that TailedCore outperforms the state-of-the-art in most settings.

Problem

• Unsupervised anomaly detection in noisy long-tail datasets
• Trade-off between noise robustness and tail class performance
• Separate handling of tail class and noise samples

Innovation

• TailSampler predicts class size via embedding similarities
• TailedCore uses noise-robust memory for anomaly detection
• Independent handling of tail class and noise samples