ABSTRACT
To accomplish efficient task offloading decisions, the authors present a DRL-based federated offloading architecture that uses multi-tiered collective learning to offload jobs. The authors offer an approach to adaptive offloading based on the multi-armed bandit theory and the offloading delay of fog nodes. An approach to offloading work is provided in the form of a game with missing information in which the existing resources are pooled and shared. In this research, the authors use a SAC (soft actor critic)-based, model-free version of deep reinforcement learning to maximise entropy and expected utility. Keep in mind that the transmission delay, a crucial component of the overall delay that they aim to minimise, considers the channel capacity. They refer to a subset of resident time optimization that, by selecting the best action from the given action space, maximises reward across multiple repetitions.
