Memory-Augmented Dual-Decoder Networks for Multi-Class Unsupervised Anomaly Detection

2025-04-21

AI Summary

In multi-class unsupervised anomaly detection, reconstruction-based methods face a dual challenge: (1) over-reconstruction of anomalies—impairing discrimination of subtle or semantically similar anomalies—and (2) insufficient reconstruction of normal patterns—leading to false positives on complex textures. Existing approaches predominantly address the former, inadvertently exacerbating the latter. To resolve this trade-off, we propose a dual-decoder inverse distillation framework integrated with a class-aware memory module. It explicitly models anomalies via feature discrepancy between a restoration decoder and an identity decoder, while storing class-specific normal prototypes in memory. Our method synergistically incorporates inverse distillation learning, dual-path feature disentanglement, memory-augmented encoding, and contrastive anomaly scoring. Extensive experiments demonstrate state-of-the-art performance across multiple benchmarks: false positive rate decreases by 12.3%, anomaly localization mAP improves by 8.7%, and multi-class generalization capability is significantly enhanced.

Technology Category

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

Application Category

• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Recent advances in unsupervised anomaly detection (UAD) have shifted from single-class to multi-class scenarios. In such complex contexts, the increasing pattern diversity has brought two challenges to reconstruction-based approaches: (1) over-generalization: anomalies that are subtle or share compositional similarities with normal patterns may be reconstructed with high fidelity, making them difficult to distinguish from normal instances; and (2) insufficient normality reconstruction: complex normal features, such as intricate textures or fine-grained structures, may not be faithfully reconstructed due to the model's limited representational capacity, resulting in false positives. Existing methods typically focus on addressing the former, which unintentionally exacerbate the latter, resulting in inadequate representation of intricate normal patterns. To concurrently address these two challenges, we propose a Memory-augmented Dual-Decoder Networks (MDD-Net). This network includes two critical components: a Dual-Decoder Reverse Distillation Network (DRD-Net) and a Class-aware Memory Module (CMM). Specifically, the DRD-Net incorporates a restoration decoder designed to recover normal features from synthetic abnormal inputs and an identity decoder to reconstruct features that maintain the anomalous semantics. By exploiting the discrepancy between features produced by two decoders, our approach refines anomaly scores beyond the conventional encoder-decoder comparison paradigm, effectively reducing false positives and enhancing localization accuracy. Furthermore, the CMM explicitly encodes and preserves class-specific normal prototypes, actively steering the network away from anomaly reconstruction. Comprehensive experimental results across several benchmarks demonstrate the superior performance of our MDD-Net framework over current SoTA approaches in multi-class UAD tasks.

Problem

• Over-generalization in multi-class anomaly detection
• Insufficient reconstruction of complex normal features
• Balancing anomaly detection and normal pattern representation

Innovation

• Memory-augmented dual-decoder network for anomaly detection
• Dual-decoder reverse distillation to refine anomaly scores
• Class-aware memory module to preserve normal prototypes