Pretraining Strategy for Neural Potentials

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Zhang, Zehua, Li, Zijie, Farimani, Amir Barati
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866916293734563840
author Zhang, Zehua
Li, Zijie
Farimani, Amir Barati
author_facet Zhang, Zehua
Li, Zijie
Farimani, Amir Barati
contents We propose a mask pretraining method for Graph Neural Networks (GNNs) to improve their performance on fitting potential energy surfaces, particularly in water systems. GNNs are pretrained by recovering spatial information related to masked-out atoms from molecules, then transferred and finetuned on atomic forcefields. Through such pretraining, GNNs learn meaningful prior about structural and underlying physical information of molecule systems that are useful for downstream tasks. From comprehensive experiments and ablation studies, we show that the proposed method improves the accuracy and convergence speed compared to GNNs trained from scratch or using other pretraining techniques such as denoising. On the other hand, our pretraining method is suitable for both energy-centric and force-centric GNNs. This approach showcases its potential to enhance the performance and data efficiency of GNNs in fitting molecular force fields.
format Preprint
id arxiv_https___arxiv_org_abs_2402_15921
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Pretraining Strategy for Neural Potentials
Zhang, Zehua
Li, Zijie
Farimani, Amir Barati
Machine Learning
Chemical Physics
We propose a mask pretraining method for Graph Neural Networks (GNNs) to improve their performance on fitting potential energy surfaces, particularly in water systems. GNNs are pretrained by recovering spatial information related to masked-out atoms from molecules, then transferred and finetuned on atomic forcefields. Through such pretraining, GNNs learn meaningful prior about structural and underlying physical information of molecule systems that are useful for downstream tasks. From comprehensive experiments and ablation studies, we show that the proposed method improves the accuracy and convergence speed compared to GNNs trained from scratch or using other pretraining techniques such as denoising. On the other hand, our pretraining method is suitable for both energy-centric and force-centric GNNs. This approach showcases its potential to enhance the performance and data efficiency of GNNs in fitting molecular force fields.
title Pretraining Strategy for Neural Potentials
topic Machine Learning
Chemical Physics
url https://arxiv.org/abs/2402.15921