Uncertainty management and reduction of machine learning potential

Akash Singh, Yumeng Li

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


Machine learning potential has drawn more and more attention in the recent years due to its ability in enabling automatic construction of high-fidelity force field potentials. They allow the successful application of reliable high throughput atomistic simulations in accelerating material discovery and design with novel functions and properties. Those methods circumvent the parameterization problem as faced by most empirical potentials and decrease model bias by using machine learning techniques to interpolate the highly nonlinear potential energy surface based on first principle calculations. Despite the advantages such as a well automated process and high flexibility of model functions, systematic uncertainty quantifications are still missing to develop uncertainty management techniques for the quantification and reduction of uncertainty involved in the developed machine learning interpolated atomic potential energy surface. In this paper, the uncertainty resources associated with the development of machine learning potential will be identified. Corresponding uncertainty quantification will be conducted through a systematic sensitivity analysis to investigate the reliability of the developed machine learning potential for graphene, an emergent promising 2D material.

Original languageEnglish (US)
Title of host publicationAIAA Scitech 2021 Forum
PublisherAmerican Institute of Aeronautics and Astronautics Inc, AIAA
Number of pages9
ISBN (Print)9781624106095
StatePublished - 2021
EventAIAA Science and Technology Forum and Exposition, AIAA SciTech Forum 2021 - Virtual, Online
Duration: Jan 11 2021Jan 15 2021

Publication series

NameAIAA Scitech 2021 Forum


ConferenceAIAA Science and Technology Forum and Exposition, AIAA SciTech Forum 2021
CityVirtual, Online

ASJC Scopus subject areas

  • Aerospace Engineering


Dive into the research topics of 'Uncertainty management and reduction of machine learning potential'. Together they form a unique fingerprint.

Cite this