A neural population view to understand and restore movement

The analysis of neural population activity across brain cortices has consistently uncovered low-dimensional subspaces that capture a significant fraction of neural variability. These “neural manifolds” are spanned by specific population-wide patterns of neural covariation. My work exploits this theoretical framework to understand the principles of robust and flexible neural computation. In this seminar, I will discuss how the neural manifold framework may help understand how the brain learns and controls behavior, and allow us to develop better Brain-Computer Interfaces (BCIs) that restore loss function after neurological injury or disease. I will start by reviewing three recent studies in which my colleagues and I adopted the neural manifold framework to describe novel computational aspects of motor control and short-term learning. First, we found that population activity spanning many different behaviours lies within a well-preserved neural manifold, with very similar dynamics across the tasks. This result was surprising given the complex changes in single neuron activity across the behaviours. Second, we studied how animals can learn to rapidly adapt their movements, even after single errors. Contrary to the prevailing view that learning must be associated with synaptic plasticity, we showed that it could occur simply through population-wide computations within a stable neural manifold. Lastly, while animals can consistently execute a well-learned behaviour, yet the neural basis for such consistency has been elusive. We found that neural population dynamics in three different cortical areas critical for movement planning, movement execution, and feedback processing remains stable for a given behaviour over days, months or even years. In the final part of my talk, I will describe how these findings can be exploited to develop a BCI-controlled neuroprosthesis to restore hand use after paralysis over extended periods of time. This fully-wireless neuroprosthesis predicts the intended activation of many paralyzed arm and hand muscles, and directly achieves the desired activation by injecting electrical current into the corresponding muscles, thereby restoring volitional control of the hand. Building on our neuroscience work, I will show how muscle activity during a broad range of naturalistic behaviors can be predicted based on the population dynamics within a stable neural manifold. Given that these population dynamics can be recovered even in the face of changing recorded neurons, these combined observations will potentially enable BCI-based neuroprostheses that are intuitive and stable over unprecedented periods of time.

Bio
Juan Álvaro Gallego is a “Talent Attraction” Postdoctoral Fellow working in the Neural and Cognitive Engineering Group at the Spanish National Research Council (CSIC). He studies how the brain learns and controls movement based on a combination of high-yield neural population recordings, computational methods, brain-computer interfaces, and behavioural analysis. He has pursued these questions in human patients, nonhuman primates, and rodents, while he worked or visited at institutions such as CSIC, Aalborg University, Northwestern University and Janelia Research Centre. Dr. Gallego’s goal is to understand how behaviours are learned, recalled, executed, and adapted, and use this knowledge to engineer neuroprosthetics to restore movement after neurological injury or disease. Throughout his career, he has presented his work in over twenty conferences and invited seminars, and authored or co-authored over seventy peer-reviewed publications, including twenty-two articles in scientific journals across a breadth of disciplines, such as biomedical engineering, neuroscience, and cybernetics.

Video transmission using zoom : https://epfl.zoom.us/j/9946495775