Computational Neuroscience: Information Processing in Neural Networks

Our goal is to understand how neural systems sense, represent, compute and learn in realistic settings, using theoretical and computational tools from statistical physics, information theory and dynamical systems theory.

In addition to uncovering the neural mechanisms underlying information processing in the brain, we seek to identify fundamental similarities and differences between biological and artificial neural networks. These insights may reveal new principles for designing more efficient, robust and biologically informed AI systems.

Our current research focuses on three interconnected areas:

• Dynamics of neural circuits: We study the dynamical and statistical properties of neural networks underlying representation, computation and learning, with an emphasis on how collective dynamics give rise to information processing.
• Statistical physics of deep learning: We use ideas from statistical physics to understand how artificial neural networks learn, generalize, and develop representations and compare these processes with learning and computation in biological neural systems.
• Energetic costs of biological computation: We investigate how physical constraints, particularly energy consumption and dissipation, shape neural computation and learning.

We work closely with the Statistical Biophysics Group at the Center for Computational Biology, which is led by Yuhai Tu, also our principal investigator. This interdisciplinary collaboration brings together perspectives from theoretical physics, molecular/cellular biology, computational neuroscience and machine learning.

Learn more about specific ongoing and past projects.