Adam Johnson
Wednesday 2nd July 2014
Time: 4pm
Basement Seminar Room
Alexandra House, 17 Queen Square, London, WC1N 3AR
A hierarchical Bayesian approach to hippocampal based schema learning
and exploration
The hippocampus plays a critical role in spatial look-ahead,
single-trial learning, memory consolidation, and imagination. Each of
these learning dynamics depends on memory schemas. Using a hierarchical
Bayesian approach, we propose a computational definition for memory
schemas as dynamic hyperparameter mixtures used to predict future task
observations. We show how this approach can account for the hippocampus
dependence of single trial learning and variable time memory
consolidation. We next show how this approach can account for
spontaneous object exploration behavior. Finally, we discuss the links
between this approach and POMDP approaches that emphasize the role of
memory in problem solving rather than memory as purely information
storage.