Upcoming Paper
My collaborator at the Center for Systems Biology and I are in the process of writing a paper on a Reinforcement Learning system which incorporates the results of Agent Spaces and the conclusions of our investigation into the relationship between Evolution Strategies and finite differences methods. In this paper we describe several distinct innovations. First, we employ the results of our paper on ES and FD to create a fully parallelizable optimization method which can proceed without any waste of computational resources or idling. Second, we demonstrate the efficacy of the agent space with a modified version of Novelty Search as an exploration method.
Our method in action. In this environment, the reward of an episode is the x-coordinate of the player at the final frame. The black box is a border which the player cannot cross. Each path in this visualization is created using the parameters of the locus agent during an epoch of optimization.