A Contrastive Learning-Guided Confident Meta-learning for Zero Shot Anomaly Detection

2025-08-25

AI Summary

To address data scarcity, high annotation costs, and dynamic environmental shifts in industrial and medical anomaly detection, this paper proposes CoZAD, a zero-shot anomaly detection framework. Methodologically, CoZAD integrates confidence-aware meta-learning with contrastive feature representation: it employs IQR-based uncertainty quantification and covariance regularization to weight learning of prototypical normal patterns while preserving boundary samples; contrastive learning constructs a discriminative feature space, enabling rapid domain adaptation without vision-language alignment or model ensembling. Innovatively combining soft-confidence learning, MAML, and feature clustering, CoZAD significantly enhances generalization to unseen anomalies. Evaluated on ten industrial and medical datasets, it achieves state-of-the-art performance: surpassing prior methods on 6 out of 7 industrial benchmarks, and attaining image-level AUROC scores of 99.2% on DTD-Synthetic and 97.2% on BTAD, alongside a pixel-level AUROC of 96.3% on MVTec-AD.

Technology Category

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

Application Category

• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Industrial and medical anomaly detection faces critical challenges from data scarcity and prohibitive annotation costs, particularly in evolving manufacturing and healthcare settings. To address this, we propose CoZAD, a novel zero-shot anomaly detection framework that integrates soft confident learning with meta-learning and contrastive feature representation. Unlike traditional confident learning that discards uncertain samples, our method assigns confidence-based weights to all training data, preserving boundary information while emphasizing prototypical normal patterns. The framework quantifies data uncertainty through IQR-based thresholding and model uncertainty via covariance based regularization within a Model-Agnostic Meta-Learning. Contrastive learning creates discriminative feature spaces where normal patterns form compact clusters, enabling rapid domain adaptation. Comprehensive evaluation across 10 datasets spanning industrial and medical domains demonstrates state-of-the-art performance, outperforming existing methods on 6 out of 7 industrial benchmarks with notable improvements on texture-rich datasets (99.2% I-AUROC on DTD-Synthetic, 97.2% on BTAD) and pixellevel localization (96.3% P-AUROC on MVTec-AD). The framework eliminates dependence on vision-language alignments or model ensembles, making it valuable for resourceconstrained environments requiring rapid deployment.

Problem

• Zero-shot anomaly detection with data scarcity challenges
• Addressing uncertainty in training samples via confidence weighting
• Creating discriminative feature spaces for rapid domain adaptation

Innovation

• Confidence-based weighting for all training data
• IQR and covariance-based uncertainty quantification
• Contrastive learning for discriminative feature spaces