Neural plasticity in decision making and memory formation

• Main Author: Garvert, M. M.
• Other Authors: Dolan, R. J. ; Behrens, T. E. J.
• Published: University College London (University of London) 2016
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.746235

Goal-directed behaviour is characterized by an ability to make inferences without direct experience. This requires a model of the environment and of ourselves, which is flexibly adjusted in light of new incoming information. This thesis uses representational functional magnetic resonance imaging (fMRI) techniques in combination with computational modelling to investigate (1) whether humans can construct models of other people’s preferences and whether this process influences their own value representation, and (2) how statistical relationships between discrete, non-spatial objects are combined into a model of the world. The first part of the thesis investigates how subjective values are computed in an intertemporal choice paradigm, and how these value computations are updated as a consequence of learning about the preferences of another. Critically, subjects’ own preferences shift towards those of the other when learning about their choices, suggesting that subjects incorporate new knowledge about others into a model of their own preferences. The underlying mechanism involves prediction errors, which introduce plasticity into subjects’ mPFC value representations, in turn resulting in a shift in subjects’ own preferences. The second part of this thesis investigates how relationships between arbitrary objects are represented in the brain. Relational knowledge is often considered analogous to spatial reasoning, where relationships are encoded in a hippocampal-entorhinal ‘cognitive map’. Here, I show that maps can also be extracted from the entorhinal cortex for discrete relationships between arbitrary stimuli, and in the absence of conscious knowledge. The representation of abstract knowledge in map-like structures suggests that inferences do not need to rely on direct experiences but can be computed anew from mapped knowledge. Together, these studies reveal how world models are represented and updated at the level of neural representations, providing a bridge between representational codes and cognitive computations.