Reinforcement Learning

Over the last two decades, the model-based approach to analysing functional magnetic resonance imaging (fMRI) data has been adopted across the cognitive neurosciences to study how computations are implemented in the brain. In this time, methods have advanced along both computational modelling and neuroimaging domains. This chapter aims to provide an introduction to the general method of integrating computational models into fMRI analyses as well as a discussion on contemporary considerations regarding these recent advances. The chapter begins with an exposition to the formalisation of qualitative psychological hypotheses into quantitatively testable and falsifiable computational models. We use examples from the conditioning and reinforcement learning literature to ground this discussion given the origin of model-based fMRI in uncovering neural correlates of learning processes. We then provide an overview of the methodological approach underlying model-based fMRI. This extends to pragmatic considerations when working in this domain with an eye towards more recent developments in both fMRI and computational modelling, such as multivariate analyses and assessing model quality, respectively. Finally, we provide examples in which computations described in the first section of the chapter were successfully bridged with fMRI analyses to provide a richer understanding of reinforcement learning in the brain. This chapter is therefore aimed at both the cognitive neuroscientist seeking to adapt computational approaches to their neuroimaging research as well as those specifically interested in learning and decision-making across levels of analyses.