Multi-Modal Foundation Models for Unified Log, Trace, and Metric Reasoning in Service Graph Diagnosis
American Journal of Machine Learning · 2025
DOI: 10.71465/ajml3451 ↗Kurz gefasst
Fachartikel in der Zeitschrift American Journal of Machine Learning, 2025. Verfasst von Adrien Moreau und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Modern cloud-native microservice architectures generate heterogeneous telemetry data including logs, distributed traces, and performance metrics, creating significant challenges for effective service graph diagnosis and root cause analysis. Traditional approaches typically process these modalities in isolation, missing critical cross-modal correlations essential for comprehensive system understanding. This paper presents a novel framework leveraging multi-modal foundation models to enable unified reasoning across logs, traces, and metrics for enhanced service graph diagnosis. We propose an architecture that integrates transformer-based foundation models with graph neural networks to capture both sequential patterns within individual modalities and structural dependencies across service graphs. The framework employs contrastive learning to align representations from different telemetry sources while preserving modality-specific characteristics crucial for accurate fault localization. Experimental evaluation on industry-standard microservice benchmarks demonstrates that our approach achieves superior performance in anomaly detection and root cause identification compared to single-modality and traditional multi-modal methods.
Zitieren
APA
Moreau, A., & Jović, K. (2025). Multi-Modal Foundation Models for Unified Log, Trace, and Metric Reasoning in Service Graph Diagnosis. American Journal of Machine Learning. https://doi.org/10.71465/ajml3451
Vancouver
Moreau A, Jović K. Multi-Modal Foundation Models for Unified Log, Trace, and Metric Reasoning in Service Graph Diagnosis. American Journal of Machine Learning. 2025. doi:10.71465/ajml3451
BibTeX
@article{Moreau2025,
author = {Moreau, Adrien and Jović, Katarina},
title = {Multi-Modal Foundation Models for Unified Log, Trace, and Metric Reasoning in Service Graph Diagnosis},
journal = {American Journal of Machine Learning},
year = {2025},
doi = {10.71465/ajml3451}
}
EUVIMED