Tsukada Lab. Computer Network Group
← Publications Int'l conference

Fast-Adapting V2X Networks: A Dynamic Context Clustering Multi-Armed Bandits

Weiqi Chi, Xiaoyang He, Manabu Tsukada

IEEE Vehicular Networking Conference (VNC 2026), Montreal, Canada, Jun 8, 2026

Summary in three lines

  • DCC-SMAB combines contextual bandits with dynamic clustering for vehicle association.
  • Vehicles reuse estimates for similar contexts and receive prior cluster knowledge when joining the network.
  • The reported comparison with UCB showed lower regret and higher network rate across simulated deployments.

Figures from the paper

Figure 1: Dynamic context clustering architecture
Figure 1. Dynamic context clustering architectureSource: Figure 1 from Chi et al., IEEE Vehicular Networking Conference (VNC 2026) (2026) (authors’ figure from the manuscript)The diagram shows how contextual observations support vehicle association decisions through shared learning.

Abstract

Only the chairs can edit While contextual multi-armed bandit (CMAB) effectively addresses non-stationary reward for problems such as millimeter-wave (mmWave) Vehicle-to-Everything (V2X) user association (UA), it suffers from slow convergence when dealing with a large-scale context space. This paper proposes a Dynamic Context Clustering Semi-distributed Multi-Armed Bandit (DCC-SMAB) algorithm for the UA problem in mmWave vehicular networks that dynamically clusters vehicle positions as contexts, thereby accelerating convergence in reward estimation without requiring additional channel state information (CSI) compared to the CMAB framework with fixed context partitioning. The algorithm exhibits strong capability in identifying mmWave channel variations, enabling rapid adaptation to non-stationary communication environments for both newly arriving and existing vehicles in the network. Extensive simulations demonstrate over 61% and 76% in average regret reduction and network transmission rate improvement compared to benchmarks using Upper Confidence Bound (UCB) policy, with robustness validated across 40 trials under varying network topologies.

How to cite

W. Chi, X. He, and M. Tsukada, “Fast-Adapting V2X Networks: A Dynamic Context Clustering Multi-Armed Bandits,” in IEEE Vehicular Networking Conference (VNC 2026), Montreal, Canada, 2026, doi: 10.1109/VNC69225.2026.11629174.

BibTeX

@inproceedings{Chi2026,
  title         = {Fast-Adapting V2X Networks: A Dynamic Context Clustering Multi-Armed Bandits},
  author        = {Weiqi Chi and Xiaoyang He and Manabu Tsukada},
  doi           = {10.1109/VNC69225.2026.11629174},
  year          = {2026},
  date          = {2026-06-08},
  urldate       = {2026-06-08},
  booktitle     = {IEEE Vehicular Networking Conference (VNC 2026)},
  address       = {Montreal, Canada},
  abstract      = {Only the chairs can edit While contextual multi-armed bandit (CMAB) effectively addresses non-stationary reward for problems such as millimeter-wave (mmWave) Vehicle-to-Everything (V2X) user association (UA), it suffers from slow convergence when dealing with a large-scale context space. This paper proposes a Dynamic Context Clustering Semi-distributed Multi-Armed Bandit (DCC-SMAB) algorithm for the UA problem in mmWave vehicular networks that dynamically clusters vehicle positions as contexts, thereby accelerating convergence in reward estimation without requiring additional channel state information (CSI) compared to the CMAB framework with fixed context partitioning. The algorithm exhibits strong capability in identifying mmWave channel variations, enabling rapid adaptation to non-stationary communication environments for both newly arriving and existing vehicles in the network. Extensive simulations demonstrate over 61% and 76% in average regret reduction and network transmission rate improvement compared to benchmarks using Upper Confidence Bound (UCB) policy, with robustness validated across 40 trials under varying network topologies.},
  keywords      = {ITS},
  pubstate      = {published},
  tppubtype     = {inproceedings}
}