Abstract : In this paper, we propose a novel approach for parallel task computation in dynamic fog computing networks, leveraging digital twin technology and distributed contextual bandit learning. Our method, Digital Twin-aided Distributed Contextual Bandit Learning (DT-DCBL), utilizes digital twins to create real-time virtual representations of fog network components, capturing their dynamic states and behaviors. This enables accurate modeling of network conditions, device capabilities, and task workloads. We employ Thompson Sampling within the contextual bandit framework to address the exploration-exploitation dilemma, efficiently learning the uncertain environment of the fog network. To optimize task offloading, we incorporate matching theory, ensuring effective pairing between task nodes and helper nodes based on their individual preferences and capabilities. By combining these advanced techniques, DT-DCBL facilitates adaptive and efficient task allocation, significantly reducing task execution latency and enhancing resource utilization in fog computing networks. Our simulation results demonstrate that the proposed approach outperforms traditional methods, achieving superior performance in terms of delay reduction and computational efficiency in dynamic fog environments.
Index terms : Digital Twin, Fog Computing, Computation Offloading, Context Bandit Learning