Multi-Agent Network Management With Dynamic Entropy-Driven Logistic Trust Aggregation

  • Jang, Minhyeok; 
  • Ben-Othman, Jalel; 
  • Chun, Hyunchae; 
  • Cho, Sungrae; 
  • Kim, Hyunbum
Citations

WEB OF SCIENCE

0
Citations

SCOPUS

0

초록

Autonomous network management increasingly fuses multiple heterogeneous detectors-such as the intrusion detectors that monitor different traffic planes for 6G and Internet of Things (IoT) security-through adaptive trust-weighted consensus. When trust is updated online, however, such systems face a fundamental stability-agility trade-off: they are either calm but slow to react to novel threats, or fast but erratic under routine noise. We identify and formalize the resulting failure modes of trust collapse and blind conformity, and propose DELTA (Dynamic Entropy-driven Logistic Trust Aggregation), a self-regulating trust-management framework. DELTA couples a Fixed-Share Redistribution regularizer, which guarantees a minimum trust quota for every detector, with an entropy-amplified logistic controller whose learning rate is driven by the current leader's error rate and amplified by the ensemble's structural entropy; this keeps the system quiescent under normal traffic yet triggers a rapid, bounded re-calibration the moment the trusted detector begins to fail. We prove that DELTA enforces a strictly positive diversity floor-making trust collapse provably impossible-and derive bounds on its transition latency and stationary volatility. Across an extensive evaluation-including robustness to delayed, missing, and adversarial feedback, comparison against expert-advice, Bayesian, and change-point baselines with confidence intervals, and validation on the real UNSW-NB15 intrusion dataset-DELTA recovers from zero-day regime shifts where naive baselines collapse below chance, while remaining an order of magnitude more stable than aggressive adaptive methods, all at O(N) computational and communication cost.

키워드

Entropy; Modeling; Management; Detectors; Labeling; Learning (artificial intelligence); Poles and zeros; Stability; Accuracy; Error analysis; Network management; distributed intrusion detection; multi-agent trust aggregation; concept drift; stability-agility trade-off; entropy-driven adaptation
제목
Multi-Agent Network Management With Dynamic Entropy-Driven Logistic Trust Aggregation
저자
Jang, Minhyeok; Ben-Othman, Jalel; Chun, Hyunchae; Cho, Sungrae; Kim, Hyunbum
DOI
10.1109/TNSM.2026.3731203
발행일
2026
유형
Article
저널명
IEEE Transactions on Network and Service Management
권
23
페이지
7525 ~ 7536