A Combinatorial Bandit Approach to UAV-aided Edge Computing

• Main Authors: Wu, Bochun, Chen, Tianyi, Wang, Xin
• Format: Conference Proceeding
• Language: English
• Subjects: Benchmark testing; Clustering algorithms; combinatorial multi-armed bandit; Computers; Delays; edge-cloud coordination; mobile edge computing; Resource management; Servers; task offloading; unmanned aerial vehicle; Unmanned aerial vehicles
• ISSN: 2576-2303
• DOI: 10.1109/IEEECONF51394.2020.9443306

The development of Internet of Things (IoT) leads to an exponential growth of computing demands. To meet the increasing demand of computation under limited resource, tasks at terminals are often offloaded to nearby mobile edge computing (MEC) nodes or remote cloud servers in order to reduce latency. It has been established that when serving as a small base station, an unmanned aerial vehicle (UAV) outperforms its terrestrial counterparts in many ways. We present in this paper a scenario, where a cluster of UAVs floating in the sky can serve multiple UEs as edge computing servers or edge relaying nodes. We aim to minimize the total delay of the offloaded tasks over time, subject to the unknown network states and the multi-agent matching conflicts. To this end, we formulate the multi-agent offloading allocation as a combinatorial multi-armed bandit (C-MAB) problem and propose an efficient algorithm. The algorithm achieves (i) an exploration/exploitation trade- off, (ii) an edge-cloud coordination, and (iii) an asymptotically optimal set of agent-action matching pairs. We prove that the proposed algorithm has a sublinear regret bound against the optimal benchmark with full a-priori knowledge. Numerical results demonstrate the merits of the proposed schemes.