Hippocampal Model of Rat Spatial Abilities Using
Temporal Difference Learning
David Foster Richard Morris Peter Dayan
Hippocampus, 10 :1-16.
Abstract
We present a model of the hippocampus dependent performance of rats in
a delayed match-to-place spatial watermaze learning task. Real rats
show slow learning in the first few days of training; but ultimately
exhibit one-trial learning. We model the early performance using a
temporal difference (TD) reinforcement learning architecture, using
place cells in the hippocampus as the representation of space. We
model late performance using a coordinate learning system which also
employs TD learning and place cells. We show a simple integration of
these two navigation models.