Bayesian approaches to Learning and Decision Mak- ing

Behavioural phenomena are central to psychiatric disorders. Computational modelling allows the learning and decision-making processes underlying behaviour to be modelled in great detail. By doing so, specific and possibly highly complex hypotheses about the underlying processes can be directly tested on the data. The first part of this chapter introduces Markov Decision Problems (MDPs) as a formal framework for decision-making. It then describes several solutions to MDPs including reinforcement learning and dynamic programming, and briefly introduces some of their key characteristics. The second part of the chapter provides a tutorial overview over how to use MDPs in a generative modelling framework to test hypotheses about learning and decision-making. The final part of the chapter discusses the methods using a few worked examples from the literature.