Canonical circuits for cerebellar learning

This proposal aims to understand how the neural circuits of the cerebellum implement computations that drive learning. The cerebellum has long been proposed to evaluate predictions about the consequences of actions using error signals delivered by the climbing fiber, a form of supervised learning. Our recent discovery that the cerebellum also exhibits signals associated with reward has transformed our view of cerebellar function and suggests that the cerebellum may also implement reinforcement learning. We will identify the sources of cerebellar reward and error signals and examine how they work together during learning in cerebellar and downstream circuits using an unprecedented combination of tools: circuit-wide and brain-wide recordings of activity using Neuropixels probes, anatomical tracing of input and target structures, and "all-optical" interrogation using 2-photon imaging and 2-photon optogenetics to provide causal links between activity in functionally defined cerebellar microzones and behaviour. These experiments will reveal how the cerebellar cortex interacts with other brain areas during learning; they will provide crucial constraints for constructing models of cerebellar cortex; and they will reveal clear targets for manipulation of an important neural circuit that may ultimately have translational relevance, particularly for treating the disorders of movement and cognition that may have cerebellar origins.