Learning to use Working Memory in Partially Observable Environments Through Dopaminergic Reinforcement. Michael Todd, Yael Niv, Jonathan Cohen

Talk: Tuesday, 10:50am Poster: Tuesday, T78
A new RL model defines working memory as a limited POMDP solver. Midbrain DA trains striatal Actor/Critic to control updating of prefrontal cortex, maximizing discounted future reward. Simulations demonstrate the model's potential to explain psychological phenomena.