The advent of neuroimaging techniques has driven advances in how we understand where in the brain different aspects of cognition are instantiated, and how this neural activity relates to behaviour. While the translation of this approach to study neuropsychiatric disorders has had some successes, it could be argued that it fails to capture what the brain is doing. Computational models serve as a bridge from brain to behaviour (see Figure 5.7.1), permitting the formulation of mechanistic hypotheses about neural computations and how they might be different in clinical conditions. Most applications of computational models to psychiatric disorders concern altered learning about the world. While many formal models of learning exist, two have had widespread success in their application to psychiatry: reinforcement learning and Bayesian models. Both models are concerned with how we learn from past experiences to form expectations about the world around us.