Studying sensorimotor integration from the perspective of neural dynamics and artificial intelligence
The CCN Group, headed by Dr. Mario Senden, is part of the Department of Cognitive Neuroscience at Maastricht University and studies the cortical perception-action loop from an interdisciplinary perspective. Our research aims to understand how the brain processes sensory information and generates appropriate motor outputs through the sensorimotor loop. Our work is built on three pillars: goal-driven deep learning, biophysical modeling and data-driven model discovery.
Goal-driven deep learning is a key aspect of our research, where we aim to develop computational models that can perform tasks in a goal-directed manner, similar to how humans and animals process information. In addition to traditional deep learning methods, we also utilize goal-driven reinforcement learning in conjunction with simulated bodies and environments. This approach enables us to uncover the neurocomputational mechanisms by which the brain controls complex actions in naturalistic environments.
Biophysical modeling is an integral part of our research and provides us with a deeper understanding of the underlying neural mechanisms. We develop models at several spatiotemporal scales, ranging from spiking neurons to population dynamics, to simulate the behavior of neural systems. Our biophysical models are based on detailed descriptions of the biophysical properties of neurons and how they interact with each other. By integrating these models with our goal-driven deep learning approach, we can learn how the brain performs complex tasks and generate predictions about how these mechanisms change with different sensory inputs and task demands. This interdisciplinary approach, combining biophysics and machine learning, enables us to gain new insights into the cortex and the sensorimotor loop
Finally, our data-driven model discovery approach focuses on using advanced techniques such as dynamic mode decomposition, sparse identification of nonlinear dynamics, time-delay embeddings and Koopman operators to identify patterns and relationships in large amounts of behavioral and neural data. These techniques allow us to uncover the underlying structure of complex data and develop models that better explain the neural mechanisms of the perception-action loop.
The CCN Group, headed by Dr. Mario Senden, is part of the Department of Cognitive Neuroscience at Maastricht University and studies the cortical perception-action loop from an interdisciplinary perspective. Our research aims to understand how the brain processes sensory information and generates appropriate motor outputs through the sensorimotor loop. Our work is built on three pillars: goal-driven deep learning, biophysical modeling and data-driven model discovery.
Goal-driven deep learning is a key aspect of our research, where we aim to develop computational models that can perform tasks in a goal-directed manner, similar to how humans and animals process information. In addition to traditional deep learning methods, we also utilize goal-driven reinforcement learning in conjunction with simulated bodies and environments. This approach enables us to uncover the neurocomputational mechanisms by which the brain controls complex actions in naturalistic environments.
Biophysical modeling is an integral part of our research and provides us with a deeper understanding of the underlying neural mechanisms. We develop models at several spatiotemporal scales, ranging from spiking neurons to population dynamics, to simulate the behavior of neural systems. Our biophysical models are based on detailed descriptions of the biophysical properties of neurons and how they interact with each other. By integrating these models with our goal-driven deep learning approach, we can learn how the brain performs complex tasks and generate predictions about how these mechanisms change with different sensory inputs and task demands. This interdisciplinary approach, combining biophysics and machine learning, enables us to gain new insights into the cortex and the sensorimotor loop
Finally, our data-driven model discovery approach focuses on using advanced techniques such as dynamic mode decomposition, sparse identification of nonlinear dynamics, time-delay embeddings and Koopman operators to identify patterns and relationships in large amounts of behavioral and neural data. These techniques allow us to uncover the underlying structure of complex data and develop models that better explain the neural mechanisms of the perception-action loop.
Contact
For the possibility of internships or lab visits, please contact Mario Senden via mario.senden@maastrichtuniversity.nl
Department of Cognitive Neuroscience
Maastricht University
Oxfordlaan 55
6229EV Maastricht
