Frontiers in Chemical Engineering

Title: Learning physical dynamics with generative machine learning

Abstract: Recent advances in large-scale scientific datasets are creating new opportunities for machine learning (ML) methods to more effectively capture scientific phenomena with greater accuracy and reach. In this talk, I will discuss how these advances are both shifting ML design paradigms and enabling new scientific inquiries. This includes investigations into developing a broader understanding of whether neural networks can autonomously discover fundamental physical relationships from data, and how this understanding informs more flexible modeling design choices that capture physical dynamics across multiple scales. I will then discuss how post-training techniques are starting to make it practical to model systems at the scales needed to converge properties that can be compared against experiment. In a similar spirit, I will also explore how generative modeling approaches grounded in statistical mechanics can accelerate the sampling of transition pathways, and as a framework to align and bridge the gap between numerically simulated data and experimental observations.

Bio: Aditi Krishnapriyan is an Assistant Professor at UC Berkeley, where she is part of Chemical and Biomolecular Engineering, Electrical Engineering and Computer Sciences, and Berkeley AI Research, as well as a faculty scientist in the Applied Mathematics division at Lawrence Berkeley National Laboratory. She holds a PhD from Stanford University, supported by the DOE Computational Science Graduate Fellowship, and was the Luis W. Alvarez Fellow in Computing Sciences at LBNL. Her research focuses on developing physics-inspired machine learning methods that bridge machine learning with physical science applications to capture phenomena across multiple length and timescales, work supported by a DOE Early Career Award and an NSF CAREER Award.