Title: From Error to Inference: Population Dynamics of Cerebellar Internal Models

Abstract:

The ability of the nervous system to identify external sensory stimuli (i.e., exafference) and distinguish them from self-generated sensory inputs (i.e., reafference) is essential for perceptual equilibrium and accurate motor control. The vestibular system provides a powerful model of this computation by selectively suppressing reafferent signals during active self-motion while preserving sensitivity to passive inputs in the vestibular nuclei—a process known as reafference cancellation. The cerebellum plays a central role in implementing this computation through internal models, with populations of Purkinje cells collectively predicting the sensory consequences of movement and generating the signals required for reafference cancellation. However, how these internal models are dynamically updated in response to changing sensory statistics remains unknown.

In this project, we introduce a novel, population-level computational framework for cerebellar learning, focusing on how internal models are updated in the cerebellum during natural behavior. By combining high-density electrophysiological recordings, controlled vestibular and proprioceptive perturbations, and data-driven modeling of neural population dynamics, we will determine how the cerebellum detects sensory prediction errors, adapts to shifting sensory feedback statistics, and flexibly implements multiple internal models across behavioral contexts.

Our central hypothesis is that cerebellar learning emerges from structured dynamics at the level of Purkinje cell populations, enabling both continuous optimization (learning) and rapid context-dependent switching (inference). This moves beyond single-neuron descriptions to establish a dynamical-system view of cerebellar computation, linking neural population activity to behavior through latent-state and model-based analyses.

The computational principles uncovered here are expected to generalize across motor and cognitive domains, providing a unifying framework for how the brain integrates sensory and motor information and updates internal models in dynamic environments. By linking neural population dynamics to adaptive behavior in a primate model, this project advances a new paradigm for studying brain-wide learning and may inform approaches to diagnosing and treating cerebellar dysfunction.