Voting-Based Multiagent Reinforcement Learning for Intelligent IoT

The recent success of single-agent reinforcement learning (RL) in Internet Things (IoT) systems motivates the study multiagent RL (MARL), which is more challenging but useful large-scale IoT. In this article, we consider a voting-based MARL problem, agents vote to make group decisions and goal maximize globally averaged returns. To end, formulate problem based on linear programming form policy optimization propose primal-dual algorithm obtain optimal solution. We also voting mechanism through distributed achieves same sublinear convergence rate as centralized learning. other words, decision making does not slow down process achieving global consensus optimality. Finally, verify our proposed with numerical simulations conduct case studies practical IoT systems.