When Unsupervised Domain Adaptation meets One-class Anomaly Detection: Addressing the Two-fold Unsupervised Curse by Leveraging Anomaly Scarcity

2025-02-28

AI Summary

This paper addresses the significant performance degradation in unsupervised anomaly detection (UAD) caused by domain shift. We propose the first fully unsupervised framework unifying domain adaptation and anomaly detection—termed UAD-DA—specifically designed for UAD settings where conventional unsupervised domain adaptation (UDA) methods fail due to a “dual unsupervised dilemma”: the source domain contains only normal samples, while the target domain is entirely unlabeled and exhibits extremely sparse anomalies. Leveraging the prior of anomaly sparsity, our method employs unsupervised clustering to identify the dominant normal cluster in the target domain and jointly aligns it with source-domain normal features in a hyperspherical embedding space. The approach integrates deep feature extraction, Deep SVDD, K-means clustering, and cross-domain alignment. Evaluated on multiple UAD domain-shift benchmarks, it achieves an average AUROC improvement of over 8% against state-of-the-art methods. The code will be publicly released.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Data Mining & Knowledge Management: Anomaly/Outlier Detection
• Computer Vision: Adversarial Attacks & Robustness

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text

Abstract

This paper introduces the first fully unsupervised domain adaptation (UDA) framework for unsupervised anomaly detection (UAD). The performance of UAD techniques degrades significantly in the presence of a domain shift, difficult to avoid in a real-world setting. While UDA has contributed to solving this issue in binary and multi-class classification, such a strategy is ill-posed in UAD. This might be explained by the unsupervised nature of the two tasks, namely, domain adaptation and anomaly detection. Herein, we first formulate this problem that we call the two-fold unsupervised curse. Then, we propose a pioneering solution to this curse, considered intractable so far, by assuming that anomalies are rare. Specifically, we leverage clustering techniques to identify a dominant cluster in the target feature space. Posed as the normal cluster, the latter is aligned with the source normal features. Concretely, given a one-class source set and an unlabeled target set composed mostly of normal data and some anomalies, we fit the source features within a hypersphere while jointly aligning them with the features of the dominant cluster from the target set. The paper provides extensive experiments and analysis on common adaptation benchmarks for anomaly detection, demonstrating the relevance of both the newly introduced paradigm and the proposed approach. The code will be made publicly available.

Problem

• Addressing domain shift in unsupervised anomaly detection.
• Leveraging anomaly scarcity to solve the two-fold unsupervised curse.
• Aligning source and target normal features using clustering techniques.

Innovation

• Unsupervised domain adaptation for anomaly detection
• Clustering to identify dominant normal cluster
• Hypersphere fitting for source-target feature alignment