Distributed Influence-Augmented Local Simulators for Parallel MARL in Large Networked Systems

Neural Information Processing Systems 

Due to its high sample complexity, simulation is, as of today, critical for the successful application of reinforcement learning.