The Evolution of Reinforcement Learning

In both the natural and artificial realms, evolution and reinforcement learning are parallel adaptive processes that work on different scales but with similar feedback mechanisms. This makes combined and coordinated study of these phenomena synergistic. In the natural world, evolution is antecedent to all forms of learning but can, nonetheless, be influenced by them. In artificial intelligence, evolutionary computation and reinforcement learning were initially developed independently and in parallel but numerous studies of their interactions exist. As reinforcement learning gains prominence in machine learning, and as it is integrated into more complex learning systems such as deep neural networks, greater understanding of how it has and could be evolved becomes increasingly relevant. This survey covers major trends in the evolution of reinforcement learning and suggests important directions for future research.