A Survey on Diffusion Models for Anomaly Detection

TL;DR

This survey examines diffusion models for anomaly detection (DMAD), framing the problem around modeling normal data distributions to identify deviations in complex, high-dimensional data. It introduces a taxonomy dividing DMAD methods into reconstruction-based, density-based, and hybrid approaches, and covers four task modalities—image, time series, video, and multimodal—along with representative methods, datasets, and evaluation metrics. Key contributions include a comprehensive taxonomy, cross-domain synthesis of techniques, discussion of challenges such as computational cost, interpretability, and robustness, and a curated set of resources and benchmarks for researchers and practitioners. The work highlights DMAD's potential to leverage sharp generative samples and precise density estimation to improve anomaly detection in real-world, heterogeneous settings, while outlining practical directions such as edge-cloud collaboration and integration with large language models.

Abstract

Diffusion models (DMs) have emerged as a powerful class of generative AI models, showing remarkable potential in anomaly detection (AD) tasks across various domains, such as cybersecurity, fraud detection, healthcare, and manufacturing. The intersection of these two fields, termed diffusion models for anomaly detection (DMAD), offers promising solutions for identifying deviations in increasingly complex and high-dimensional data. In this survey, we review recent advances in DMAD research. We begin by presenting the fundamental concepts of AD and DMs, followed by a comprehensive analysis of classic DM architectures including DDPMs, DDIMs, and Score SDEs. We further categorize existing DMAD methods into reconstruction-based, density-based, and hybrid approaches, providing detailed examinations of their methodological innovations. We also explore the diverse tasks across different data modalities, encompassing image, time series, video, and multimodal data analysis. Furthermore, we discuss critical challenges and emerging research directions, including computational efficiency, model interpretability, robustness enhancement, edge-cloud collaboration, and integration with large language models. The collection of DMAD research papers and resources is available at https://github.com/fdjingliu/DMAD.

Figures (3)

• Figure 1: Taxonomy of diffusion models for anomaly detection.
• Figure 2: Pipeline illustration of reconstruction-based AD methods: (a) basic, (b) latent space, and (c) conditional reconstruction.
• Figure 3: Pipeline illustration of density-based AD methods: (a) score function-based and (b) diffusion time estimation method.