Deep Reinforcement Learning for Task Offloading in Mobile Edge Computing Systems

In mobile edge computing systems, an node may have a high load when large number of devices offload their tasks to it. Those offloaded experience processing delay or even be dropped deadlines expire. Due the uncertain dynamics at nodes, it is challenging for each device determine its offloading decision (i.e., whether not, and which should task to) in decentralized manner. this work, we consider non-divisible delay-sensitive as well dynamics, formulate problem minimize expected long-term cost. We propose model-free deep reinforcement learning-based distributed algorithm, where can without knowing models other devices. To improve estimation cost incorporate long short-term memory (LSTM), dueling Q-network (DQN), double-DQN techniques. Simulation results show that our proposed algorithm better exploit capacities nodes significantly reduce ratio average compared with several existing algorithms.