Program

A number of recent projects involve the collection of behavioural data from large numbers of subjects in carefully controlled cognitive circumstances. These offer an unprecedented opportunity to examine hitherto hidden aspects of the acquisition and maintenance of performance by individuals. I will describe two projects in which we have built descriptive models of the performance of more than 100 mice on a simple perceptual decision-making task. In the first, I will discuss our analysis of the long-run shaping and learning of the mice, from their very first choice to expert behaviour. This uses a non-parametric Bayesian model to separate the fast and slow changes that jointly characterize acquisition. In the second project, I will describe our use of a ladder of increasingly flexible and complex additions to a simple behavioural model in order to capture rich details of the asymptotic performance of the subjects. This shows subtle, but important, gaps in our original account.