Pathologies of mood are at the root of many psychiatric disorders. In this talk, I will present recent theoretical and empirical work that uses the framework of reinforcement learning to understand mood in both health and disease. I will first introduce a family of computational models in which mood is conceptualized as a leaky integral of the advantage of an agent’s actions, and will provide a normative derivation of this theory as an analogue of momentum in stochastic gradient descent. I will describe how these models might account for psychiatric phenomena such as persistent negative mood and mood volatility, and will present data from a behavioral task investigating the association between mood and learning in major depression and bipolar disorder.
Back to Computational Psychiatry