Talk title: A first-principles approach to ML and applications to quantum materials

Abstract: Machine learning (ML) and AI hold much promise for the prediction and targeted design of material properties. They offer new paradigms for combining insights developed from theory, computation, and experiment and for bridging the microscopic atomistic world with the macroscopic. To progress in this direction, it is important to understand how models built with deep neural networks work. Drawing from statistical physics ideas and tools, I will present findings from a body of work in which my collaborators and I developed scientific foundations for deep learning. I will discuss new connections we have discovered between large-width deep neural networks, Gaussian processes, and kernels; the emergence of linear models during training and phase transitions away from them; and first-principles guidelines for scaling up deep neural networks. These insights are likely to guide the design of ML for materials. I will discuss early stages of a second research program using ML/AI to advance the field of materials research. I will describe ongoing work using large language models to perform analytic Hartree-Fock mean-field calculations for a particular class of quantum materials. Taken together, the simultaneous pursuit of scientific understanding of ML/AI with the rich application frontiers in the physics and chemistry of materials can unlock qualitatively new scientific possibilities for discovery and design.

Yasaman Bahri is a Research Scientist at Google DeepMind working at the intersection of the physical sciences and machine learning. She completed her Ph.D. in Physics at UC Berkeley as an NSF Fellow, specializing in the theory of condensed matter. Her doctoral work investigated new quantum phases of matter through the themes of symmetry, topology, and localization. She is a past recipient of the Rising Stars Award in EECS and has been an invited lecturer at the Les Houches School of Physics. She has multidisciplinary interests at the confluence of machine learning and AI, quantum materials, and statistical physics.

Website: https://sites.google.com/view/yasamanbahri/home

Event Date
-
Location
Tan 775A
Event ID
237002