Fast Adaptive Task Offloading in Edge Computing Based on Meta Reinforcement Learning

Multi-access edge computing (MEC) aims to extend cloud service the network reduce traffic and latency. A fundamental problem in MEC is how efficiently offload heterogeneous tasks of mobile applications from user equipment (UE) hosts. Recently, many deep reinforcement learning (DRL) based methods have been proposed learn offloading policies through interacting with environment that consists UE, wireless channels, However, these weak adaptability new environments because they low sample efficiency need full retraining updated for environments. To overcome this weakness, we propose a task method on meta learning, which can adapt fast small number gradient updates samples. We model as Directed Acyclic Graphs (DAGs) policy by custom sequence-to-sequence (seq2seq) neural network. train seq2seq network, synergizes first order approximation clipped surrogate objective. The experimental results demonstrate latency up 25% compared three baselines while being able