Title: Neural Integration in Motor Control with Graph-Driven Dynamics

The integration of sensory stimuli, such as proprioceptive and vestibular inputs, with internal motor commands is crucial for robust postural response. Particularly, maintenance of stable posture in response to unexpected vestibular stimuli relies on head-stabilizing vestibulo-spinal reflexes, which require suppression during intentional head movements. Despite recent progress in motor control research, significant unknowns persist as to the precise mechanisms underlying how the brain determines the need and degree of such suppression. New technologies that enable simultaneous recordings from hundreds of neurons across trials and conditions present an opportunity to discover these mechanisms, however the complexity of these new rich datasets requires developing advanced computational methods.Current computational tools including low-dimensional simple projections and linear dynamical models, are often too simplistic to capture the complexities of large-scale neural data, while 'black box' machine learning models, though powerful in their representational capacity, tend to be uninterpretable. Here, we propose a computational framework to uncover the latent neural computational units underlying movement control in both the cerebellum and the cerebrum, aiming to understand how the brain integrates diverse sources of information to perform robust postural responses