A cerebellar population coding model for sensorimotor learning.

The cerebellum plays a critical role in sensorimotor learning, and in particular using error information to keep the sensorimotor system well-calibrated. Here we present a population-coding model of how the cerebellum compensates for motor errors. The model consists of a two-layer network, one corresponding to the cerebellar cortex and the other to the deep cerebellum nuclei, where the units within each layer are tuned to two features, the direction of the movement and the direction of the error. We evaluated our model through a series of behavioral experiments that test sensorimotor adaptation across a wide range of perturbation schedules. The model successfully accounts for interference from prior learning, the effects of error uncertainties, and learning in response to perturbations that vary across different time scales. Importantly, the model does not require any modulation of the parameters or context-dependent processes during adaptation. Our results provide a novel framework to understand how context and environmental uncertainty modulate cerebellar-dependent learning.