Big Data for Physical Science: Statistical Learning and Model
Reduction of Complex Chemical Reaction Networks
Qian Yang, Stanford University
Throughout the course of human history, progress in the physical sciences has evolved from experimentation, to the development of theoretical laws, to in more recent decades advanced computer simulation. In this talk, I will advocate that the natural and necessary next step in this evolution is data-??driven discovery. This paradigm is relatively new for the physical sciences, but is already showing promise in areas from battery design to combustion modeling, and represents a field ripe for development and innovation.
As an example of this data-??driven paradigm in action, I will present our recent work that seeks to automatically learn complex chemistry from expensive physics simulations that currently require weeks to perform on the world’s fastest supercomputers. These simulations are computed around the world by scientists in a variety of fields, and I will show that it is possible to exploit this existing data using statistical learning to build fast, interpretable, and predictive models of chemistry that can then be used to rapidly simulate related but previously unexplored chemical systems. Additionally, systems of interest in the physical sciences are often extremely complex and difficult for human interpretation, and I will describe our recent work developing an efficient, data-??driven, L1-??regularization based algorithm for model reduction of nonlinear dynamical systems. Finally, I will describe some open problems and challenges that are especially important to scientific machine learning, highlighting opportunities for a path forward in this exciting space.
Bio:
Qian Yang is a postdoctoral scholar in the Materials Computation and Theory Group at Stanford advised by Evan Reed. Her research interests lie at the intersection of machine learning, computational science, and the physical sciences. Qian recently completed her Ph.D. from the Institute for Computational and Mathematical Engineering at Stanford University, where she was a 2016-??17 Accel Innovation Scholar, and holds a B.A. in applied mathematics/computer science from Harvard College. Prior to graduate school, she worked in industry as a software developer, and lead software development for a start-??up using machine learning techniques to build a diagnostic device for balance disorders in the elderly.