Past Event: Oden Institute Seminar

How to model interatomic interactions with machine learning

Alexey Kolmogorov, Associate Professor, Binghamton University, State University of New York

Abstract

**This seminar will be presented LIVE in POB 6.304.  It will also be streamed live via Zoom.**

Machine learning interatomic potentials (MLIPs) have emerged as attractive alternatives to traditional potentials for simulation of large systems. Being general and flexible approximators, they can be tuned to describe diverse atomic configurations with near ab initio accuracy. I will overview the recent advances in the field that have made the construction of reliable MLIPs possible and demonstrate their performance in structure prediction. I will also describe an automated framework for generating representative structure datasets and a stratified scheme for training neural network models for systems with several chemical elements. These approaches implemented in our MAISE package have allowed us to generate a library of MLIPs for a number of metals and alloys and make first predictions of synthesizable compounds.

Biography

Aleksey Kolmogorov is an Associate Professor of Physics at Binghamton University, State University of New York. After receiving Physics degrees from Moscow Institute of Physics and Technology (MS, 1995) and Penn State University (PhD, 2004), he worked in Materials Departments at Duke University (2004-2007) and the University of Oxford (2008-2012). With background in Physics, Materials Science, and Computer Science, he develops and uses materials modeling tools at the intersection of the three disciplines. His group focuses on the prediction of new superconducting, energy storage, and topological materials with density functional theory and machine learning methods.