MECAD: A multi-expert architecture for continual anomaly detection

2025-12-17

AI Summary

To address knowledge degradation in anomaly detection caused by continual evolution of product categories in industrial settings, this paper proposes a multi-expert continual learning framework. Methodologically: (1) we introduce a novel dynamic expert assignment mechanism based on feature similarity, routing each new class to the most semantically proximate expert; (2) we design a lightweight knowledge preservation strategy integrating optimized core-set selection with task-specific replay buffers, eliminating the need for full retraining. Evaluated on the MVTec AD dataset comprising 15 object categories, our approach achieves a mean AUROC of 0.8259—significantly outperforming single-expert baselines—while effectively mitigating catastrophic forgetting. The framework strikes a favorable trade-off among detection accuracy, inference efficiency, and scalability to evolving category distributions.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Computer Vision: Object Detection & Categorization
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

In this paper we propose MECAD, a novel approach for continual anomaly detection using a multi-expert architecture. Our system dynamically assigns experts to object classes based on feature similarity and employs efficient memory management to preserve the knowledge of previously seen classes. By leveraging an optimized coreset selection and a specialized replay buffer mechanism, we enable incremental learning without requiring full model retraining. Our experimental evaluation on the MVTec AD dataset demonstrates that the optimal 5-expert configuration achieves an average AUROC of 0.8259 across 15 diverse object categories while significantly reducing knowledge degradation compared to single-expert approaches. This framework balances computational efficiency, specialized knowledge retention, and adaptability, making it well-suited for industrial environments with evolving product types.

Problem

• Continual anomaly detection for evolving object classes
• Dynamic expert assignment based on feature similarity
• Incremental learning without full model retraining

Innovation

• Multi-expert architecture for continual anomaly detection
• Dynamic expert assignment based on feature similarity
• Coreset selection and replay buffer for incremental learning