Reinforcement Learning and Hippocampal Dynamics

Recent experimental findings on hippocampal representational dynamics such as route replay and sweeps match intuitive notions from reinforcement learning including transiently representing potential trajectories and reward locations. We explore these intuitions within a formal reinforcement learning framework and examine how these representational dynamics might be integrated with reinforcement learning algorithms. We suggest that hippocampal representational dynamics can be best integrated within a model-based reinforcement learning framework and show how this framework can be used to cultivate specific quantitative predictions for the control processes that direct and utilize hippocampal representations.