Keywords: Multi-agent Systems; Reinforcement Learning; Deep Learning; Deep Reinforcement Learning; Actor-Critic methods; Partial Observability; Scalability; Evolution Strategies

Abstract: This paper is a review of literature on multi-agent reinforcement and deep learning, including a high-level overview on a popular and modern approach of reinforcement learning, called the actor-critic method. It discusses how scalability can be a challenge for reinforcement learning, and how can various methods, including the actor-critic method, help resolve this challenge. In the process, it presents us with various applications and evaluation/analysis work carried out in the field, that would showcase reinforcement learning’s potential to solve real-world problems. Given the above, the paper also briefly discusses Evolution Strategies (ES), a recent technique that has gained traction in the artificial intelligence and machine learning communities as a potential alternative to reinforcement learning. I had also included a thorough appendix with fundamental information on Deep Learning Techniques.