Faculty Seminar Alexander Mathis : Integrating Body and Brain: Computational Approaches for Proprioception and Motor Control
Absract
Understanding how the brain controls movement represents one of neuroscience's grand challenges. Central to this challenge is proprioception, which proves crucial for bodily perception and motor control, despite remaining poorly understood. Specifically, how the nervous system integrates information from numerous distributed receptors throughout the body remains an open question. We leveraged the principle that body and brain evolved as an integrated system to develop an approach for reverse-engineering proprioception. We expressed competing hypotheses of the proprioceptive system as quantifiable computational tasks. We found that neural networks optimized to perform these computational tasks develop representations that generalize to predict neural dynamics in the brain. Validating this approach further, these models reproduced classic proprioceptive illusions without any explicit training on illusion data. Motor control presents another core body-brain integration challenge: how does the nervous system coordinate hundreds of muscles and joints for skilled behavior? To understand motor control, we leveraged emerging fast simulators and took inspiration from the theory of motor learning to develop policies capable of controlling high-dimensional, nonlinear biomechanical models for skilled tasks. Analyzing these models revealed human-like low-dimensional control spaces and insights into principles of sensorimotor control.
Bio
Alexander Mathis is an Assistant Professor at EPFL whose research group works at the intersection of computational neuroscience and machine learning. His work focuses on quantifying behavior and investigating how the brain generates movement, while developing accessible open-source software tools such as DeepLabCut. He completed his doctorate at Ludwig Maximilian University Munich (LMU) in 2012 after studying pure mathematics. He subsequently held postdoctoral positions at Harvard University and the University of Tübingen (as a Marie Skłodowska-Curie Postdoctoral Fellow). With his students, he won consecutive MyoChallenges at NeurIPS (2022 and 2023). His work has been recognized with awards, including the 2024 Robert Bing Prize, the 2023 Eric Kandel Young Neuroscientists Prize, and the 2023 Frontiers of Science Award. He is a member of the newly formed Simons Collaboration on Ecological Neuroscience (SCENE), which investigates how environmental affordances shape neural representations.
This seminar is part of the evaluation of Prof. Alexander Mathis for the promotion to Associate Professor.
Understanding how the brain controls movement represents one of neuroscience's grand challenges. Central to this challenge is proprioception, which proves crucial for bodily perception and motor control, despite remaining poorly understood. Specifically, how the nervous system integrates information from numerous distributed receptors throughout the body remains an open question. We leveraged the principle that body and brain evolved as an integrated system to develop an approach for reverse-engineering proprioception. We expressed competing hypotheses of the proprioceptive system as quantifiable computational tasks. We found that neural networks optimized to perform these computational tasks develop representations that generalize to predict neural dynamics in the brain. Validating this approach further, these models reproduced classic proprioceptive illusions without any explicit training on illusion data. Motor control presents another core body-brain integration challenge: how does the nervous system coordinate hundreds of muscles and joints for skilled behavior? To understand motor control, we leveraged emerging fast simulators and took inspiration from the theory of motor learning to develop policies capable of controlling high-dimensional, nonlinear biomechanical models for skilled tasks. Analyzing these models revealed human-like low-dimensional control spaces and insights into principles of sensorimotor control.
Bio
Alexander Mathis is an Assistant Professor at EPFL whose research group works at the intersection of computational neuroscience and machine learning. His work focuses on quantifying behavior and investigating how the brain generates movement, while developing accessible open-source software tools such as DeepLabCut. He completed his doctorate at Ludwig Maximilian University Munich (LMU) in 2012 after studying pure mathematics. He subsequently held postdoctoral positions at Harvard University and the University of Tübingen (as a Marie Skłodowska-Curie Postdoctoral Fellow). With his students, he won consecutive MyoChallenges at NeurIPS (2022 and 2023). His work has been recognized with awards, including the 2024 Robert Bing Prize, the 2023 Eric Kandel Young Neuroscientists Prize, and the 2023 Frontiers of Science Award. He is a member of the newly formed Simons Collaboration on Ecological Neuroscience (SCENE), which investigates how environmental affordances shape neural representations.
This seminar is part of the evaluation of Prof. Alexander Mathis for the promotion to Associate Professor.
Practical information
- Informed public
- Free
Organizer
- Deanship School of Life Sciences
Contact
- Manuelle Mary