de Cothi, William John;
(2020)
Predictive maps in rats and humans for spatial navigation.
Doctoral thesis (Ph.D), UCL (University College London).
Preview |
Text
W_deCothi_thesis_with_corrections.pdf Download (9MB) | Preview |
Abstract
The ability to navigate space is an essential part of mammalian life. Over the last 50 years, much research has investigated on how the mammalian brain represents space in the activity of populations of neurons, particularly focussing upon cells in the hippocampal formulation. But how does the brain integrate these representations to guide flexible and efficient navigational decision making? A useful way to approach this question is from the field of reinforcement learning, which seeks to address how an agent should act in its environment in order to maximise some form of reward signal. Typically solutions to a reinforcement learning problem are split into a dichotomy of model-free and model-based approaches. Here we investigate the biological validity of an intermediary approach called the successor representation, which works by forming a predictive map of the environment. First, we compare these three reinforcement learning methods to rat and human behaviour on a transition revaluation spatial navigation task, and show that the biological behaviour is most similar to that of a successor representation agent. Then we propose a neurally plausible implementation of the successor representation, based upon a set of known neurobiological features - boundary vector cells. We show that the place and grid cells generated using this model provide a good account of biological data for a variety of environmental manipulations, including dimensional stretches, barrier insertions, and the influence of environmental geometry on the hippocampal representation of space.
Archive Staff Only
 |
View Item |
