Anomaly detection using Diffusion-based methods

TL;DR

This paper evaluates diffusion-based models like DDPMs and Diffusion Transformers for anomaly detection, demonstrating their superior performance, scalability, and robustness over traditional methods in complex datasets.

Contribution

It introduces the application of diffusion-based architectures to anomaly detection and benchmarks their effectiveness against traditional approaches.

Findings

Diffusion models outperform traditional methods in anomaly detection accuracy.

Reconstruction error is a key factor in improving detection performance.

Diffusion models scale effectively to high-dimensional datasets.

Abstract

This paper explores the utility of diffusion-based models for anomaly detection, focusing on their efficacy in identifying deviations in both compact and high-resolution datasets. Diffusion-based architectures, including Denoising Diffusion Probabilistic Models (DDPMs) and Diffusion Transformers (DiTs), are evaluated for their performance using reconstruction objectives. By leveraging the strengths of these models, this study benchmarks their performance against traditional anomaly detection methods such as Isolation Forests, One-Class SVMs, and COPOD. The results demonstrate the superior adaptability, scalability, and robustness of diffusion-based methods in handling complex real-world anomaly detection tasks. Key findings highlight the role of reconstruction error in enhancing detection accuracy and underscore the scalability of these models to high-dimensional datasets. Future directions include optimizing encoder-decoder architectures and exploring multi-modal datasets to further advance diffusion-based anomaly detection.