Machine Learning and Atomistic Simulations of Materials: Progresses and Challenges
Title: Machine Learning and Atomistic Simulations of Materials: Progresses and Challenges
Time: 15:10-16:10, Monday, 2021/11/29
The link for the online seminar :
Google Meet
https://meet.google.com/htj-dwhd-zpz
Abstract:
In the past few years, machine learning (ML) have drawn a significant amount of attention in almost all aspects of science, engineering, and industry. Among all these scenario of applications the computational materials science is of no exception. In this talk, I will discuss several recent applications of ML models in atomistic simulations of complex materials, and their limitations. I will discuss the energy partitioning scheme proposed by Behler and Parinello, and their applications to the microstructures of complex perovskite materials; then, I will discuss another ML model, the spectral neighbor analysis potential (SNAP) model, which are utilized in our group recently to study the large-scale (~0.5M atoms) plasticity of high entropy alloy and intercalation of lithium ions into graphite. Finally, I will briefly discuss our progress in developing a self-learning ML model based on the energy partitioning scheme, in which in principle minimum knowledge a priori is required for running the model.
