Our site uses cookies necessary for its proper functioning. To improve your experience, other cookies may be used: you can choose to disable them. This can be changed at any time via the Cookies link at the bottom of the page.


Mark Walton

Department of Experimental Psychology, University of Oxford, U.K.
Website : https://www.waltonlab.org/


Adaptive behaviour, state inference and mesolimbic dopamine


Mark Walton is a Professor of Behavioural Neuroscience at the University of Oxford.  Research in his group is focused on understanding the neural mechanisms shaping motivation and adaptive reward seeking, with a particular interest into how neurotransmitters such as dopamine regulate these processes in rodents on a moment-by-moment timescale.  To do this, they use in vivo techniques to enable fine-scale measurement and manipulation of neurochemistry and neural circuits integrated during rich behavioural tasks that can tease apart animals’ behavioural strategies and motivations.


Adaptative behavious requires learning which actions lead to desired outcomes and updating these preferences when the world changes. Reinforcement learning (RL) has provided an influential account of how this works in the brain, with reward prediction errors (RPEs) updating estimates of the values of states and/or actions, in turn driving choices.  However, it is increasing clear that an ability to infer statistical relationships and hidden states of the world also plays an important role in shaping adaptive behaviour.  Intriguingly, brain recordings have shown that not only prefrontal cortex but also the dopamine system can reflect knowledge of such hidden states.  However, this raises several conundrums: first, if state inference, not RL mediates flexible reward-guided behaviour, why does dopamine look and act like an RPE?  Conversely, if value updates driven by dopaminergic RPEs are central to flexible, how does this generate the signatures of hidden state inference seen in the data?  Here, I will present data from highly-trained mice performing a structured two-step decision task that belief updates shape dopamine responses but are not caused by them.  These data can be reconciled by a neural network model in which cortex infers hidden states by predicting observations and basal ganglia uses RL mediated by dopaminergic RPEs to learn appropriate actions.