Generalized prediction errors in the human cerebellum

Project Summary In addition to motor control and learning, the cerebellum is intimately linked to cognition. This project is designed to closely examine the cerebellum's role in nonmotor domains, namely, reinforcement learning and statistical learning. We hypothesize that the structure's core computations for sensorimotor learning can be generalized to nonmotor contexts. It is critical to understand how the cerebellum contributes to nonmotor learning – this knowledge will support the development of novel mechanistic and clinical insights into cerebellar function, and human learning in general. Foundational theoretical work has described how the cerebellum typifies an ideal substrate for supervised motor learning. This theory made testable empirical predictions that have been borne out in experiments in animals using tasks such as Pavlovian eyeblink conditioning and vestibular-ocular reflex adaptation, revealing facts about cerebellar sensorimotor processes in exquisite detail. But what about a cerebellar role in other task domains? Here we address this question. The proposed work integrates behavioral, neuroimaging, and computational techniques to develop a new framework for generalized cerebellar learning computations. The research plan centers on three Specific Aims. In Aim 1 we use computationally guided functional neuroimaging (fMRI) to examine the role of the cerebellum in reinforcement learning. We test the idea that the cerebellum processes reward predictions and prediction errors, the core computations of reinforcement learning. We also posit a constraint on cerebellar learning computations, namely that the structure only contributes to learning when the temporal interval between events is brief (i.e., subsecond). Aim 2 takes a similar approach to the domain of visual statistical learning, examining sensory predictions and prediction errors in the cerebellum and further testing the proposed timing constraint. In Aims 1-2 we also measure cerebro-cerebellar connectivity to position the cerebellum within broader task-specific learning networks, and to ask if cerebro-cerebellar connectivity covaries with behavior. In Aim 3 we examine causal contributions of the cerebellum to nonmotor learning, testing a large sample of individuals with cerebellar pathology and contrasting their behavior with matched controls. Computational analyses will be used to detect and characterize the hypothesized deficits. This project proposes a new framework for understanding the contributions of the cerebellum to nonmotor learning and will provide new insight into the broader role of the cerebellum in health and disease.