Uncertainty quantification for neural network potential foundation models

Jenna A. Bilbrey

doi:10.1038/s41524-025-01572-y

2025-04-24

DOI：10.1038/s41524-025-01572-y

原文链接

原文求助

摘要

En 中文

为了使神经网络势函数（NNPs）得到广泛应用，研究人员必须能够信任模型的输出结果。然而，神经网络的黑箱特性和其固有的随机性常常成为阻碍因素，尤其是对于在广阔化学空间上训练的基础模型。预测时提供的不确定性信息有助于减少对NNPs的抵触。本研究详细介绍了两种不确定性量化（UQ）方法。通过微调基础模型集合的读出层，读出集成（readout ensembling）提供了关于模型不确定性的信息；而通过用分布预测替代点预测，分位数回归（quantile regression）提供了关于底层训练数据中不确定性的信息。我们以MACE-MP-0模型为例验证了该方法，将UQ应用于基础模型和一系列微调模型。读出集成和分位数方法产生的不确定性被证明是衡量NNP输出质量的两种不同指标。

Keyword:

ENSEMBLES

期刊

npj Computational Materials

IF:

11.9

论文数:

2.4K

被引数:

1.7W

机构

Pacific Northwest Natl Lab

学者数:

788

论文数: 297

被引数: 91

引用论文

TorchANI: A Free and Open Source PyTorch-Based Deep Learning Implementation of the ANI Neural Network Potentials TorchANI: 基于PyTorch的ANI神经网络潜力的免费开源深度学习实现

JOURNAL OF CHEMICAL INFORMATION AND MODELING

IF5.3

2020-06-22

205

OAAI

Gao, Xiang; Ramezanghorbani, Farhad; Isayev, Olexandr; Smith, Justin S.; Roitberg, Adrian E.

Single-model uncertainty quantification in neural network potentials does not consistently outperform model ensembles 神经网络潜力中的单模型不确定性量化并不总是优于模型集合

NPJ COMPUTATIONAL MATERIALS

IF11.9

2023-12-16

OAAI

Tan, Aik Rui; Urata, Shingo; Goldman, Samuel; Dietschreit, Johannes C. B.; Gomez-Bombarelli, Rafael

Learning together: Towards foundation models for machine learning interatomic potentials with meta-learning 共同学习: 通过元学习建立机器学习的原子间势的基础模型

NPJ COMPUTATIONAL MATERIALS

IF11.9

2024-07-17

OAAI

Allen, Alice E. A.; Lubbers, Nicholas; Matin, Sakib; Smith, Justin; Messerly, Richard; Tretiak, Sergei; Barros, Kipton

Less is more: Sampling chemical space with active learning 少即多: 通过主动学习对化学空间进行采样

JOURNAL OF CHEMICAL PHYSICS

IF3.1

2018-05-22

571

OAAI

Smith, Justin S.; Nebgen, Ben; Lubbers, Nicholas; Isayev, Olexandr; Roitberg, Adrian E.

A review of uncertainty quantification in deep learning: Techniques, applications and challenges 深度学习中的不确定性量化: 技术、应用与挑战

INFORMATION FUSION

IF15.5

2021-12-01

1.2K

OAAI

Abdar, Moloud; Pourpanah, Farhad; Hussain, Sadiq; Rezazadegan, Dana; Liu, Li; Ghavamzadeh, Mohammad; Fieguth, Paul; Cao, Xiaochun; Khosravi, Abbas; Acharya, U. Rajendra; Makarenkov, Vladimir; Nahavandi, Saeid

Evidential Deep Learning for Guided Molecular Property Prediction and Discovery 用于引导分子特性预测和发现的证据深度学习

ACS CENTRAL SCIENCE

IF10.4

2021-07-27

107

OAAI

Soleimany, Ava P.; Amini, Alexander; Goldman, Samuel; Rus, Daniela; Bhatia, Sangeeta N.; Coley, Connor W.

The Open Catalyst 2022 (OC22) Dataset and Challenges for Oxide Electrocatalysts 开放催化剂2022 (OC22) 数据集和氧化物电催化剂的挑战

ACS CATALYSIS

IF13.1

2023-02-16

OAAI

Tran, Richard; Lan, Janice; Shuaibi, Muhammed; Wood, Brandon M.; Goyal, Siddharth; Das, Abhishek; Heras-Domingo, Javier; Kolluru, Adeesh; Rizvi, Ammar; Shoghi, Nima; Sriram, Anuroop; Therrien, Felix; Abed, Jehad; Voznyy, Oleksandr; Sargent, Edward H.; Ulissi, Zachary; Zitnick, C. . Lawrence

Effects of aleatoric and epistemic errors in reference data on the learnability and quality of NN-based potential energy surfaces 参考数据中偶然误差和认知误差对基于NN的势能面可学习性及质量的影响

Artificial Intelligence Chemistry

IF0

2024-06-01

OAAI

Sugata Goswami; Silvan Käser; Raymond J. Bemish; Markus Meuwly

Performance Assessment of Universal Machine Learning Interatomic Potentials: Challenges and Directions for Materials' Surfaces 通用机器学习原子间势的性能评估: 材料表面的挑战和方向

ACS APPLIED MATERIALS & INTERFACES

IF8.2

2024-07-11

OAAI

Focassio, Bruno; Freitas, Luis Paulo M.; Schleder, Gabriel R.