A Dual-Level Game-Theoretic Approach for Collaborative Learning in UAV-Assisted Heterogeneous Vehicle Networks
Zihao Ding, Jun Huang, Qiang Duan, Cheng Zhang, Yanxiao Zhao, Shuyang Gu
Abstract
Knowledge diversity and knowledge forgetting are two major issues in sustaining collaborative learning within heterogeneous vehicle networks. These issues become especially severe when vehicles possess varying sensing capabilities, computational resources, and domain expertise, leading to fragmented learning and unstable knowledge retention over time. To address these challenges, we propose a dual-level game-theoretic approach. We first formulate a new metric, Utility-of-Information (UoI), to characterize the features of knowledge learning, retention, and consolidation. Based on this metric, we design a game-theoretic dual-level approach, which comprises a lower-level coalition formation game where vehicles self-organize into “teacher-student” coalitions based on their UoI profiles, and an upper-level UAV resource allocation game where vehicle coalitions compete for limited communication resources. To optimize both levels of the game, we design a unified reinforcement learning-based framework that enables adaptive searching for optimization under dynamic network conditions. Experimental results demonstrate that our approach effectively addresses knowledge diversity and significantly mitigates the effects of knowledge forgetting in UAV-assisted heterogeneous vehicle networks.