We are a research group at University of Pennsylvania led by Prof. Nat Trask.
We develop machine learning methods grounded in the structure of physics and geometry. By combining tools from geometric mechanics, exterior calculus, and variational modeling with modern AI architectures, we create interpretable and reliable models for complex physical systems in high-consequence engineering settings. Our work spans simulation, scientific discovery, and data-driven inference across multiscale and multiphysics domains including energy, material discovery, fusion, and soft matter. Our group’s primary focus is on the construction of learning frameworks that encode physical principles by construction in neural architectures so that models provide the stability, physical realizability, and performance guarantees that are crucial to traditional modeling and simulation but lacking in contemporary machine learned models.
Check our group’s Github!
We are looking for creative and mathematically-inclined PhD students, Postdocs, and Master students to join the team (see openings) !
