Princeton
Title: Learning in a complex and high-dimensional environment
Abstract: The major reason learning is challenging is the world is complex and high dimensional, and there can be a delay between a stimulus or action and the resulting outcome. I will share two stories that reveal different ways the brain uses built-in priors to simplify this learning problem and thus help overcome these challenges. The first project concerns learning of a visual-decision making task and emphasizes the role of a feature-specific reward prediction error signal in midbrain dopamine neurons in supporting initial task acquisition. The second project focuses on conditioned taste aversion, in which an animal learns to associate a taste with much later feedback from the gut, and reveals the importance of novelty detection circuitry and reactivation by the gastric feedback of recently consumed novel flavor representations in learning across long delays.