Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster D in Poster Session D: Wednesday, August 5, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Uncertainty minimization generates hippocampus-like dynamics in learning and prediction signals in statistical learning

Pieter Barkema¹, Clare Press¹, Peter Kok¹; ¹University College London

Presenter: Pieter Barkema

The hippocampus (HC) has been implicated in learning as well as predicting perception, but how it balances the two functions is unknown. Based on a Bayesian Brain framework, we hypothesized that uncertainty minimization could unify the two modes in one system, and could generate the dynamics found in BOLD activity of HC during Statistical Learning (SL). We trained a hierarchical probability learner during SL, and compared the learned parameters to HC dynamics. We found that precision-weighted prediction errors resemble information in HC dynamics during invalidly predicted trials – decreasing while learning completes, while precision-weighted prediction errors resemble invalid trials – increasing as prediction become more certain. Taken together, we show that uncertainty minimization generates HC-like dynamics, and is a possible mechanism for combining prediction and learning in HC during SL.

Topic Area: Memory, Learning & Knowledge Structures