Computation offloading and resource allocation based on distributed deep learning and software defined mobile edge computing

In this paper, a software defined mobile edge computing (SD-MEC) in Internet of Things (IoT) is investigated, in which multiple IoT devices choose to offload their computation tasks to an appropriate edge server to support the emerging IoT applications with strict computation-intensive and latency-critical requirements. In considered SD-MEC networks, a joint computation offloading and power allocation problem is proposed to minimize the utility of weighted delay and power consumption in the distributed dense IoT. The optimization problem is a mixed-integer non-linear programming problem and difficult to solve by general optimization tools due to the nonconvexity and complexity. We propose a distributed deep learning based computation offloading and resource allocation (DDL-CORA) algorithm for SD-MEC IoT in which multiple parallel deep neural networks (DNNs) are invoked to generate the optimal offloading decision and resource scheduling. Additionally, we design a shared replay memory mechanism to effectively store newly generated offloading decisions which are further used to train and improve DNNs. The simulation results show that the proposed DDL-CORA algorithm can reduce the system utility on average 7.72% than reference Deep Q-network (DQN) algorithm and 31.9% than reference Branch-and-Bound (BNB) algorithm, and keep a good tradeoff between the complexity and utility performance.