50 min
Seminar, Hosted by Simons Institute for the Theory of Computing
Working Memory Influences Reinforcement Learning Computations in Brain and Behavior
UC Berkeley
Recording
Abstract
Anne Collins studies how working memory and reinforcement learning interact when people learn stimulus-action associations. Varying the number of associations separates rapid, capacity-limited learning from slower incremental learning. Behavioral, fMRI and EEG evidence suggests that working memory changes reward-based value updates. Computational models distinguish independent, competing and cooperative accounts, with preliminary evidence that working memory contributes to the expectation used to compute prediction errors.