Robust Unsupervised Microservice Fault Diagnosis via Masked State Representation and Directional Deviation Analysis

View at publisher Summarise with VIRAT
Authors: Yupeng Xu, Gang He
Cited by: 0
References: 20

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Reliable fault diagnosis in large-scale microservice systems requires accurate and robust diagnosis across Anomaly Detection (AD), Failure Triage (FT), and Root Cause Localization (RCL). Unsupervised unified frameworks reduce deployment and maintenance costs by sharing representations across tasks. However, they can be fragile under noisy multi-modal monitoring data. Specifically, representation learning is sensitive to hyperparameters, FT boundaries become unstable in high-dimensional deviation spaces, and RCL may be misled by propagation-induced symptom nodes. Therefore we propose Robust-ART, a robust unsupervised unified diagnosis framework that integrates a full-process solution covering Anomaly detection, failure Triage, and Root cause localization. During pre-training, we introduce a masked autoencoder (MAE) auxiliary objective that reconstructs structured masked observations, strengthening spatio-temporal dependency learning and improving stability. For FT, we replace heuristic partitioning with a von Mises–Fisher mixture model (vMF-MM) to cluster normalized deviation directions on a unit hypersphere, yielding noise-tolerant probabilistic assignments. For RCL, we further fuse directional similarity and deviation magnitude via a magnitude-weighted scoring rule to better distinguish true root causes from followers. Our evaluations on two public microservice benchmark datasets demonstrate that Robust-ART consistently outperforms current baselines in terms of AD (improving by 2.3% to 2.4%), FT (improving by 9.0% to 12.7%), and RCL (improving by 2.6% to 24.9%).

Fields of Study & Subjects

Software System Performance and Reliability Fault Detection and Control Systems Anomaly Detection Techniques and Applications

Publication details

Journal / source
International Conference on Artificial Intelligence and Social Network Systems (AISNS)
Pages
72–78
Publication date
9 January 2026
Language
EN