Robust and scalable uncertainty estimation with conformal prediction for machine-learned interatomic potentials

被引:11
|
作者
Hu, Yuge [1 ]
Musielewicz, Joseph [2 ]
Ulissi, Zachary W. [2 ]
Medford, Andrew J. [1 ]
机构
[1] Georgia Inst Technol, Dept Chem & Biomol Engn, Atlanta, GA 30332 USA
[2] Carnegie Mellon Univ, Dept Chem Engn, Pittsburgh, PA USA
来源
关键词
machine learning force fields; neural network force fields; uncertainty quantification; conformal prediction; DENSITY-FUNCTIONAL THEORY;
D O I
10.1088/2632-2153/aca7b1
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Uncertainty quantification (UQ) is important to machine learning (ML) force fields to assess the level of confidence during prediction, as ML models are not inherently physical and can therefore yield catastrophically incorrect predictions. Established a-posteriori UQ methods, including ensemble methods, the dropout method, the delta method, and various heuristic distance metrics, have limitations such as being computationally challenging for large models due to model re-training. In addition, the uncertainty estimates are often not rigorously calibrated. In this work, we propose combining the distribution-free UQ method, known as conformal prediction (CP), with the distances in the neural network's latent space to estimate the uncertainty of energies predicted by neural network force fields. We evaluate this method (CP+latent) along with other UQ methods on two essential aspects, calibration, and sharpness, and find this method to be both calibrated and sharp under the assumption of independent and identically-distributed (i.i.d.) data. We show that the method is relatively insensitive to hyperparameters selected, and test the limitations of the method when the i.i.d. assumption is violated. Finally, we demonstrate that this method can be readily applied to trained neural network force fields with traditional and graph neural network architectures to obtain estimates of uncertainty with low computational costs on a training dataset of 1 million images to showcase its scalability and portability. Incorporating the CP method with latent distances offers a calibrated, sharp and efficient strategy to estimate the uncertainty of neural network force fields. In addition, the CP approach can also function as a promising strategy for calibrating uncertainty estimated by other approaches.
引用
收藏
页数:23
相关论文
共 50 条
  • [21] A machine-learned interatomic potential for silica and its relation to empirical models
    Erhard, Linus C.
    Rohrer, Jochen
    Albe, Karsten
    Deringer, Volker L.
    [J]. NPJ COMPUTATIONAL MATERIALS, 2022, 8 (01)
  • [22] Machine-Learned Potentials by Active Learning from Organic Crystal Structure Prediction Landscapes
    Butler, Patrick W. V.
    Hafizi, Roohollah
    Day, Graeme M.
    [J]. JOURNAL OF PHYSICAL CHEMISTRY A, 2024, 128 (05): : 945 - 957
  • [23] Machine-learned prediction of the electronic fields in a crystal
    Teh, Ying Shi
    Ghosh, Swarnava
    Bhattacharya, Kaushik
    [J]. MECHANICS OF MATERIALS, 2021, 163
  • [24] Bayesian uncertainty quantification for machine-learned models in physics
    Gal, Yarin
    Koumoutsakos, Petros
    Lanusse, Francois
    Louppe, Gilles
    Papadimitriou, Costas
    [J]. NATURE REVIEWS PHYSICS, 2022, 4 (09) : 573 - 577
  • [25] A theoretical case study of the generalization of machine-learned potentials
    Wang, Yangshuai
    Patel, Shashwat
    Ortner, Christoph
    [J]. Computer Methods in Applied Mechanics and Engineering, 2024, 422
  • [26] Bayesian uncertainty quantification for machine-learned models in physics
    Yarin Gal
    Petros Koumoutsakos
    Francois Lanusse
    Gilles Louppe
    Costas Papadimitriou
    [J]. Nature Reviews Physics, 2022, 4 : 573 - 577
  • [27] A theoretical case study of the generalization of machine-learned potentials
    Wang, Yangshuai
    Patel, Shashwat
    Ortner, Christoph
    [J]. COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING, 2024, 422
  • [28] Machine-Learned Assessment and Prediction of Robust Solid State Storage System Reliability Physics
    Sarkar, Jay
    Peterson, Cory
    Sanayei, Amir
    [J]. 2018 IEEE INTERNATIONAL RELIABILITY PHYSICS SYMPOSIUM (IRPS), 2018,
  • [29] Machine learned interatomic potentials using random features
    Dhaliwal, Gurjot
    Nair, Prasanth B.
    Singh, Chandra Veer
    [J]. NPJ COMPUTATIONAL MATERIALS, 2022, 8 (01)
  • [30] Optimal data generation for machine learned interatomic potentials
    Allen, Connor
    Bartok, Albert P.
    [J]. MACHINE LEARNING-SCIENCE AND TECHNOLOGY, 2022, 3 (04):