Collaborative Computation Offloading Scheme Based on Deep Reinforcement Learning

Deep Reinforcement Learning (DRL)-based computation offloading scheme has been proposed to improve the Quality of Experience (QoE). However, the existing DRL does not consider the temporal states because of the fully connected layer. Also, DRL learns the policy regardless of the importance of experience. To solve these problems, we propose a collaborative computation offloading scheme with DRL. First, we define the objective function about task service time and load balance. Second, we utilize the Least Absolute Shrinkage and Selection Operator (LASSO) regression in the backbone network for considering temporal states. Finally, we prioritize the experience according to the Temporal Difference (TD) error and learning loss. The simulation results show that the proposed scheme achieves high QoE due to low task service time and high load balance.