EUVIMEDEUVIMED entdecken
Crossref

Agentic AI for Intelligent Network Monitoring: Autonomous Anomaly Interpretation via Bayesian Learning Automata

Jinchao Huang, Bin Yang, Chen Cheng, Pengyao Xu, Lexi Xu, Xinzhou Cheng, Chong Di

IEEE Transactions on Cognitive Communications and Networking · 2026 · 12 · 8946-8961

DOI: 10.1109/tccn.2026.3705820 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Cognitive Communications and Networking, 2026. Verfasst von Jinchao Huang und 6 weiteren Autor:innen. Die Arbeit nennt 43 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Shandong Province, Young Talent of Lifting Engineering for Science and Technology in Shandong.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Anomaly detection in modern communication networks generates alerts that require actionable interpretation for effective operational response. Existing interpretation methods rely on static metric ranking, which cannot adapt to evolving system conditions or incorporate operational feedback. This paper proposes an Agentic AI framework that reformulates anomaly interpretation as an adaptive decision process. An autonomous explanation agent operates on top of a pretrained anomaly detector, observing metric-level signals and selecting informative metric subsets through iterative environment interaction. We instantiate this framework with BLA+-Interp, an extended Bayesian Learning Automaton that maintains Beta posteriors over metric relevance and employs Thompson sampling for exploration-exploitation balance. The agent learns continuously from feedback without requiring detector retraining, achieving low computational overhead suitable for real-time deployment. Experiments on real-world multivariate time series datasets demonstrate that BLA+-Interp achieves 90% precision and 50% higher F1 score compared to static baselines, with significantly faster convergence than alternative adaptive methods (UCB1, EXP3, $\epsilon $ -greedy bandit) at operationally relevant low interaction budgets, as confirmed by paired Wilcoxon tests. Additional experiments characterise robustness to noisy, bounded short-lag delayed, partial, and non-stationary feedback as well as perturbed detector signals, and report per-anomaly latency on a production-grade workstation. Together these results establish the effectiveness, robustness, and real-time suitability of agentic approaches for interpretable network monitoring.

Förderung

  • National Natural Science Foundation of China
  • Natural Science Foundation of Shandong Province
  • Young Talent of Lifting Engineering for Science and Technology in Shandong
  • Foundation of Shandong Key Laboratory of Intelligent Technology and Equipment for Unmanned Systems
  • Key R&D Program of Shandong Province, China
  • 20 Guidelines for New Colleges in Jinan City

Zitieren

APA

Huang, J., Yang, B., Cheng, C., Xu, P., Xu, L., Cheng, X., & Di, C. (2026). Agentic AI for Intelligent Network Monitoring: Autonomous Anomaly Interpretation via Bayesian Learning Automata. IEEE Transactions on Cognitive Communications and Networking, 12, 8946-8961. https://doi.org/10.1109/tccn.2026.3705820

Vancouver

Huang J, Yang B, Cheng C, Xu P, Xu L, Cheng X, et al. Agentic AI for Intelligent Network Monitoring: Autonomous Anomaly Interpretation via Bayesian Learning Automata. IEEE Transactions on Cognitive Communications and Networking. 2026;12:8946-8961. doi:10.1109/tccn.2026.3705820

BibTeX
@article{Huang2026,
  author = {Huang, Jinchao and Yang, Bin and Cheng, Chen and Xu, Pengyao and Xu, Lexi and Cheng, Xinzhou and Di, Chong},
  title = {Agentic AI for Intelligent Network Monitoring: Autonomous Anomaly Interpretation via Bayesian Learning Automata},
  journal = {IEEE Transactions on Cognitive Communications and Networking},
  year = {2026},
  volume = {12},
  pages = {8946--8961},
  doi = {10.1109/tccn.2026.3705820}
}
Kostenlos starten →