Skip to main content
OSTI.GOVU.S. Department of Energy Office of Scientific and Technical Information
U.S. Department of Energy
Office of Scientific and Technical Information
 

Data efficiency and extrapolation trends in neural network interatomic potentials

Journal Article · · Machine Learning: Science and Technology
;

Abstract

Recently, key architectural advances have been proposed for neural network interatomic potentials (NNIPs), such as incorporating message-passing networks, equivariance, or many-body expansion terms. Although modern NNIP models exhibit small differences in test accuracy, this metric is still considered the main target when developing new NNIP architectures. In this work, we show how architectural and optimization choices influence the generalization of NNIPs, revealing trends in molecular dynamics (MD) stability, data efficiency, and loss landscapes. Using the 3BPA dataset, we uncover trends in NNIP errors and robustness to noise, showing these metrics are insufficient to predict MD stability in the high-accuracy regime. With a large-scale study on NequIP, MACE, and their optimizers, we show that our metric of loss entropy predicts out-of-distribution error and data efficiency despite being computed only on the training set. This work provides a deep learning justification for probing extrapolation and can inform the development of next-generation NNIPs.

Sponsoring Organization:
USDOE
OSTI ID:
1996854
Alternate ID(s):
OSTI ID: 2000830
Journal Information:
Machine Learning: Science and Technology, Journal Name: Machine Learning: Science and Technology Journal Issue: 3 Vol. 4; ISSN 2632-2153
Publisher:
IOP PublishingCopyright Statement
Country of Publication:
United Kingdom
Language:
English

References (42)

Constructing high-dimensional neural network potentials: A tutorial review journal March 2015
Loss landscapes and optimization in over-parameterized non-linear systems and neural networks journal July 2022
Exploring the necessary complexity of interatomic potentials journal December 2021
Neural Network Potential Energy Surfaces for Small Molecules and Reactions journal October 2020
Machine Learning Force Fields journal March 2021
Gaussian Process Regression for Materials and Molecules journal August 2021
Combining Machine Learning and Computational Chemistry for Predictive Insights Into Chemical Systems journal July 2021
Extending the Applicability of the ANI Deep Learning Molecular Potential to Sulfur and Halogens journal June 2020
Linear Atomic Cluster Expansion Force Fields for Organic Molecules: Beyond RMSE journal November 2021
Performance and Cost Assessment of Machine Learning Interatomic Potentials journal October 2019
Quantum-chemical insights from deep tensor neural networks journal January 2017
Automated discovery of a robust interatomic potential for aluminum journal February 2021
Differentiable sampling of molecular geometries with uncertainty-based adversarial attacks journal August 2021
E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials journal May 2022
A universal strategy for the creation of machine learning-based atomistic force fields journal September 2017
Recent advances and applications of machine learning in solid-state materials science journal August 2019
Deep learning study of tyrosine reveals that roaming can lead to photodamage journal June 2022
Machine learning for molecular and materials science journal July 2018
A universal graph deep learning interatomic potential for the periodic table journal November 2022
ANI-1: an extensible neural network potential with DFT accuracy at force field computational cost journal January 2017
Energy landscapes for machine learning journal January 2017
Atom-centered symmetry functions for constructing high-dimensional neural network potentials journal February 2011
Molecular dynamics with coupling to an external bath journal October 1984
Hierarchical modeling of molecular energies using a deep neural network journal June 2018
Extending the accuracy of the SNAP interatomic potential form journal June 2018
SchNet – A deep learning architecture for molecules and materials journal June 2018
Machine learning for interatomic potential models journal February 2020
FCHL revisited: Faster and more accurate quantum machine learning journal January 2020
Bayesian, frequentist, and information geometric approaches to parametric uncertainty quantification of classical empirical interatomic potentials journal June 2022
Ab initio thermodynamics of liquid and solid water journal January 2019
Shaping the learning landscape in neural networks around wide flat minima journal December 2019
Archetypal landscapes for deep neural networks journal August 2020
The atomic simulation environment—a Python library for working with atoms journal June 2017
How robust are modern graph neural network potentials in long and hot molecular dynamics simulations? journal November 2022
Atomic cluster expansion for accurate and transferable interatomic potentials journal January 2019
Metadynamics Simulations of the High-Pressure Phases of Silicon Employing a High-Dimensional Neural Network Potential journal May 2008
Gaussian Approximation Potentials: The Accuracy of Quantum Mechanics, without the Electrons journal April 2010
Deep Potential Molecular Dynamics: A Scalable Model with the Accuracy of Quantum Mechanics journal April 2018
Generalized Neural-Network Representation of High-Dimensional Potential-Energy Surfaces journal April 2007
Machine learning of accurate energy-conserving molecular force fields journal May 2017
Machine learning unifies the modeling of materials and molecules journal December 2017
Universally Sloppy Parameter Sensitivities in Systems Biology Models journal January 2007

Similar Records

E(3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials
Journal Article · Wed May 04 00:00:00 EDT 2022 · Nature Communications · OSTI ID:1866303

Related Subjects