SegDiff: Expert-Guided Segment-Wise Diffusion Model for Cloud Burst Anomaly Detection

Abstract

Effective anomaly detection is critical for the operational management of cloud infrastructure, as it is key to preventing system failures. Cloud computing systems are subject to various anomalies from factors like abnormal loads, resource contention, and malicious attacks. These anomalies typically manifest as segment-level patterns rather than isolated point anomalies. However, most existing methods are designed for point anomalies, making them ineffective for segment-level detection and unable to provide feature-level root cause diagnosis. To solve these problems, we propose SegDiff, a diffusion model guided by an expert networks at the segment level. Expert networks computes anomaly scores at the feature level to guide the diffusion model’s reverse denoising process, and furthermore, provides localization capabilities. Within our diffusion model, we introduce the TimeSegFormer module. It treats time segments as tokens to effectively capture both local and global temporal dependencies, thereby assisting in segment-level anomaly detection. Extensive experiments on four datasets demonstrate that our model surpasses state-of-the-art methods in anomaly detection accuracy.

Keywords

Cloud burst data / Anomaly detection / Feature localization / Segment-Wise diffusion model / Expert networks

RIGHTS & PERMISSIONS

Higher Education Press 2026

Just Accepted

This article has successfully passed peer review and final editorial review, and will soon enter typesetting, proofreading and other publishing processes. The currently displayed version is the accepted final manuscript. The officially published version will be updated with format, DOI and citation information upon launch. We recommend that you pay attention to subsequent journal notifications and preferentially cite the officially published version. Thank you for your support and cooperation.