Supervisors
Project Description
The human brain maintains structured neural activation in motor and somatosensory areas even in the absence of motor output, caused by injury. Rehabilitation of such injuries requires creating neural interfaces that can read out this structured activity of the brain and transform this activity into commands controlling external devices. Mastering the control of such neural interfaces requires practice and can be facilitated by providing proprioceptive feedback via neurostimulation akin to the harmonised feedback between movement and proprioceptive sensations in healthy individuals. Both decoding movement intentions from the brain and returning proprioceptive sensations via neurostimulation critically rely on identifying neural dynamics based on the electrophysiological recordings of neural population activity. However, the ability to precisely control transitions from one dynamical state to another is an unmet challenge.
The human brain maintains structured neural activation in motor and somatosensory areas even in the absence of motor output, caused by injury. Rehabilitation of such injuries requires creating neural interfaces that can read out this structured activity of the brain and transform this activity into commands controlling external devices. Mastering the control of such neural interfaces requires practice and can be facilitated by providing proprioceptive feedback via neurostimulation akin to the harmonised feedback between movement and proprioceptive sensations in healthy individuals. Both decoding movement intentions from the brain and returning proprioceptive sensations via neurostimulation critically rely on identifying neural dynamics based on the electrophysiological recordings of neural population activity. However, the ability to precisely control transitions from one dynamical state to another is an unmet challenge.
This research project aims to uncover strategies for identifying and controlling neural dynamics in motor and somatosensory areas of human cortex. The approach aims to learn input-output properties of neuronal populations from intracranial recording and stimulation in human clinical studies to enable data-driven model-based control of the neural dynamics, using deep learning and optimal control theory. By developing novel dynamical system identification methods, integrating deep generative networks with Kalman filters (DeepKalman filters), and employing optimal control strategies, the project seeks to advance our understanding of neural codes for closed-loop control of neural activity. In addition, our project seeks to investigate the hypothesis that neural dynamics are universally shared among individuals, as demonstrated in animal research. We aim to explore the potential of leveraging these shared dynamics to facilitate quicker adaptation of Brain-Computer Interfaces (BCIs) to individual users. This is an interdisciplinary project, spanning basic neuroscience and computer science, with implications for enhancing Brain-Computer Interfaces (BCIs) and progressing first-in-human clinical translation.
