Abstract:

Reinforcement-learning theories provide a normative perspective on learning and decision-making. In the 1990s, neurophysiology experiments revealed an exceptional correspondence between the activity of midbrain dopamine neurons and the reward prediction error (RPE) signal used to train computers in a reinforcement-learning algorithm called temporal difference (TD) learning. Studies of midbrain dopamine neurons play a pivotal role at the interface of empirical and theoretical studies. A theoretical framework for reinforcement
learning has facilitated the interpretation of neurophysiology data and has guided the design of future studies. Here we discuss recent developments in the interplay between experimental findings and theories of dopamine signaling. In particular, recent studies emphasize the importance of state uncertainty in the neurobiological implementation of reinforcement learning.

Publisher's Version