Jan E. Gerken

Jan E. Gerken is an Assistant Professor leading a research group focused on the mathematical foundations of artificial intelligence, supported by the Wallenberg AI, Autonomous Systems and Software Program. His work bridges theoretical physics and machine learning, with a PhD in string theory where he computed genus-one scattering amplitudes. Currently, he investigates wide neural networks, neural tangent kernels, and their connections to quantum field theory, alongside mathematical aspects of geometric deep learning and spherical computer vision.

His research spans equivariant neural networks, geometric deep learning, neural tangent kernels, and topological physics applications. Recent publications demonstrate emergent equivariance in deep ensembles, gauge-equivariant models for Chern number prediction, and HEALPix-based transformers for spherical data. Key collaborations include work on modular graph forms in string theory and diffeomorphic counterfactuals for explainable AI.

Articles highlight trends in symmetry-driven architectures, spherical data processing, and theoretical physics-motivated deep learning. Gerken’s contributions include open-source implementations like HEAL-SWIN and theoretical frameworks for equivariant training dynamics. His work addresses both foundational mathematical questions and practical applications in medical imaging, autonomous systems, and climate modeling.