Unit on Neural Computations in Learning

National Institute Of Mental Health

Investigators

Linked publications & trials

Abstract

In the last fiscal year, we have made substantial progress on multiple fronts within this research program. We have developed advanced computational models of how task structure and timing is integrated by animals into internal models of the environment to facilitate reward learning in order to understand how the brain can flexibly support reward-guided behaviors in different settings. In a recent publication, we have found that the hippocampus is critical for “chunking” task structure into blocks, allowing midbrain dopaminergic learning circuits to exploit the stability of this inferred block structure for flexible reward-guided behavior [1]. Results from two related modeling studies on the temporal component of reward prediction - in ventral striatal circuits and in dopamine reward prediction errors after exposure to cocaine – were both presented at an international conference (Multi-Disciplinary Conference on Reinforcement Learning and Decision Making). An additional study using computational modeling to tease out how engagement in a task can fluctuate and thus influence how reward learning progresses through a session, was presented at the annual Society for Neuroscience annual meeting. Together, these studies have given us a more detailed picture of how reward learning relies on the way a learner understands the task in which they are engaged, and how distinct brain areas mediate distinct aspects of these internal task representations. In collaboration, we have extended these modeling techniques to understand how learning is altered in the brain and in behavior after chronic alcohol exposure, a state known to produce critical changes to the way reward information is processed in the brain [2]. In our experimental work, we are halfway through a novel empirical study aimed at testing how reward timing and flavor impact the internal structure of reward predictions in the brain. We are now engaged in preliminary data analysis and quality control. In parallel, we have been developing a number of new analysis approaches in advance of collection of the remaining electrophysiology dataset by further analyzing previously published data. In parallel, we have partially completed data collection for a novel reward learning task for human participants that provides an analogue of the animal reward learning paradigms we study. This study gives us translational leverage on how individual differences in timing ability relate to the temporal component of reward prediction and learning, and how each relate to self-reported measures of symptoms that are clinically-relevant for disorders of mental health in these large online cohorts.