How does the brain combine historical knowledge and online processing in decision making?

I will investigate how people‘s brains combine different sources of information to decide which action to make. In particular, I will look at how people combine new information with what they know from experience.
I will use brain imaging to see which parts of the brain ‘light up‘ when people play a computer game which requires them to combine their past experience with new information to catch falling ‘bombs‘. I will make computer simulations of what goes on in the brain when people play the game, and try to associate activity in the simulations with the activity we see in the brains of real people.
I hope to improve out understanding of how people make decisions and select actions. Although the research I am proposing here is not clinically oriented, it will contribute to an understanding of how the healthy brain processes information to make decisions: This is important for understanding many psychiatric problems, in which these processes are distorted. A further benefit to health research will be that through this project and my training I will help improve the technology we have for interpreting brain imaging data, which is used in neurological and psychiatric research.

How does the brain combine sources of predictive information when the nature of that information is qualitatively different, so that different neural structures and different computational processes are required to process it? This is the focus of the proposed research. In particular, I will look at how the brain combines information learned through a history of experience, using reinforcement learning algorithms in the basal ganglia, with information which must be processed online during a behavioural event, using forward models in the cerebellum. I hypothesise that when both types of information are important in an action selection task, they will be brought together in the Intra-Parietal Sulcus and combined in a Bayesian fashion that takes account of the relative values of the two information types, as defined by the degree of uncertainty in each.

To test this model I will develop a paradigm in which historical experience and online information processing must be combined to achieve optimal behaviour. I will use functional magnetic resonance imaging (fMRI) to investigate which brain networks are involved in processing each type of information. I will compare network models using Dynamic Causal Modelling. I will also investigate how brain activity relates to specific algorithms (as opposed to stimuli or responses) by constructing a Bayesian ‘computer participant‘ and using the parameters of its computations to model fMRI data. Finally, I will apply multivariate statistics as a measure of the presence of information about different parameters of the computational algorithm in the spatial maps of the IPS. I will learn each of these three new techniques from a named collaborator - the first two in my host laboratory, and the third during the planned year abroad. However, the combination of Baysian modelling and multivariate techniques is novel, and exciting because it may prove a very effective tool in computational fMRI.

I will also investigate the computational functioning of the cerebellum itself by applying ‘computer participant‘ approach. I will fit different models of cerebellar computations to behaviour and fMRI data from that structure. Whilst many computational models of the cerebellum are extant, and fMRI would be an ideal technique to compare them (as it allows simultaneous observation of all processing ‘nodes‘), the spatial resolution of fMRI as so far been too poor to apply this approach to the cerebellum. This will change with the installation of a high-field scanner at FMRIB, offering a unique opportunity to investigate this problem.