Collaborative Task Offloading Based on Deep Reinforcement Learning in Heterogeneous Edge Networks

The field of multimedia video has shown a vigorous development trend in recent years. The massive heterogeneous data computing tasks generated by new multimedia applications cause user experience to face two major challenges: latency and energy consumption. The emergence of multimedia edge networks allows users' computing tasks to be offloaded to edge servers for execution, which will greatly reduce communication delays. But on the contrary, multimedia edge networks also face problems such as limited communication and computing resources, unknown global system information and difficulty in multi-agent collaboration. In this paper, we first describe a multimedia edge network architecture. In this architecture, multiple edge servers use communication base stations to provide multimedia computing services to a large number of users. Then a multi-agent collaborative decision-making model is proposed, which ensures that agents with different strategies can ultimately achieve consistent decision-making. Finally, we provide a distributed computing task scheduling algorithm based on deep reinforcement learning. A large number of simulation experiments have proven the superiority of this method in minimizing delay, energy consumption and convergence.