Dr. Arghya Bhowmik
13/08/2026
David Wang Auditorium, 3rd Floor, Dalia Maydan Bldg.
13:30
This talk presents graph-based learning methods for accelerating key stages of these workflows, including equivariant graph neural networks for electron-density prediction, interatomic potentials, uncertainty-aware active learning, and direct Hessian estimation. It also explores generative and reinforcement-learning approaches for designing molecules, catalysts, crystalline and disordered materials, with emphasis on data-efficient discovery beyond conventional foundation models. Together, these developments illustrate how graph-based surrogate and generative models can enable faster, more targeted exploration of complex materials spaces.