DImensionality-Reduced Encoded Clusters with Stratified (DIRECT) Sampling for Robust Training of Machine Learning Interatomic Potentials

Explore the innovative DImensionality-Reduced Encoded Clusters with sTratified (DIRECT) sampling approach for robust training of Machine Learning Interatomic Potentials (MLIPs) in this comprehensive seminar. Delve into the importance of MLIPs in enabling accurate simulations of materials at scales beyond conventional first-principles approaches. Discover how DIRECT sampling selects a robust training set of structures from a large and complex configuration space, using the Materials Project relaxation trajectories dataset with over one million structures and 89 elements. Learn about the development of an improved materials 3-body graph network (M3GNet) universal potential that extrapolates more reliably to unseen structures. Examine the use of molecular dynamics (MD) simulations with universal potentials like M3GNet as a rapid alternative to expensive ab initio MD for creating large configuration spaces for target materials systems. Gain insights into the application of this scheme combined with DIRECT sampling to develop a reliable moment tensor potential for titanium hydrides without iterative augmentation of training structures.

Ji Qi: DImensionality-Reduced Encoded Clusters with sTratified (DIRECT) sampling for Robust ...