 |
Center for Nonlinear Studies |
I will present an overview of current challenges, opportunities and strategies in the problem of optimizing Density Functional Theory (DFT) methods using machine learning techniques. I will present three separate but interconnected strategies to expand DFT methodologies into the realm of neural network and modern machine learning techniques. First is the traditional force field approach, where neural networks are replacing expensive DFT calculations while promising same accuracy. Secondly, I will present our efforts on optimizing the exchange and correlation functional aiming to improve simultaneously the densities and total energies in the Kohn-Sham Hamiltonian. I will show where results are satisfactory and where results are still behind traditional XC functionals. Finally I will introduce our work towards developing a universal electron force field, which aims to achieve multi scaling capabilities while incorporating, at a semi-classical level ions and electrons on the same footing.
About the Speaker: Marivi Fernandez-Serra is a professor in the Department of Physics & Astronomy at Stony Brook University, and is affiliated with the Institute for Advanced Computational Science at Stony Brook. She obtained her Ph.D. in 2005 from the University of Cambridge. Following this she was a postdoctoral researcher at CECAM in Lyon, France. She joined the faculty at Stony Brook University in 2008, and was awarded the DOE Early Career Award in 2010. Her research has spanned from investigations of the electronic structure and dynamical properties of water and solids in aqueous-solvated environments to ab-initio modeling of solids for dark-matter detection. Her current focus is to leverage traditional electronic structure methods with modern machine learning techniques to improve the quantum mechanical description of properties of water and related nonequilibrium phenomena.
+1-415-655-0002 US Toll
Access code: 26342282832
Host: Vidushi Sharma and Alexander J. White (T-1)