Uncertainty quantification of artificial neural network based machine learning potentials

Yumeng Li, Weirong Xiao, Pingfeng Wang

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Atomistic simulations play an important role in the material analysis and design by being rooted in the accurate first principles methods that free from empirical parameters and phenomenological models. However, successful applications of MD simulations largely depend on the availability of efficient and accurate force field potentials used for describing the interatomic interactions. As a powerful tool revolutionizing many areas in science and technology, machine learning techniques have gained growing attentions in the field of material science and engineering due to their potentials to accelerate the material discovery through their applications in surrogate model assisted material design. Despite tremendous advantages of employing machine learning techniques for the development of force field potentials as compared to conventional approaches, the uncertainty involved in the machine learning interpolated atomic potential energy surface has not drew much attention although it is an important issue. In this paper, the uncertainty quantification study is performed for the machine learning interpolated atomic potentials, and applied to the titanium dioxide (TiO2), an industrially relevant and well-studies material. The study results indicated that quantifying uncertainties is an indispensable task that must be performed along with the atomistic simulation process for a successful application of the machine learning based force field potentials.

Original languageEnglish (US)
Title of host publicationMaterials
Subtitle of host publicationGenetics to Structures
PublisherAmerican Society of Mechanical Engineers (ASME)
ISBN (Electronic)9780791852170
DOIs
StatePublished - 2018
EventASME 2018 International Mechanical Engineering Congress and Exposition, IMECE 2018 - Pittsburgh, United States
Duration: Nov 9 2018Nov 15 2018

Publication series

NameASME International Mechanical Engineering Congress and Exposition, Proceedings (IMECE)
Volume12

Conference

ConferenceASME 2018 International Mechanical Engineering Congress and Exposition, IMECE 2018
CountryUnited States
CityPittsburgh
Period11/9/1811/15/18

ASJC Scopus subject areas

  • Mechanical Engineering

Fingerprint Dive into the research topics of 'Uncertainty quantification of artificial neural network based machine learning potentials'. Together they form a unique fingerprint.

Cite this