Physical Encoding Improves OOD Performance in Deep Learning Materials Property Prediction

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Fu, Nihang, Omee, Sadman Sadeed, Hu, Jianjun
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917729803436032
author Fu, Nihang
Omee, Sadman Sadeed
Hu, Jianjun
author_facet Fu, Nihang
Omee, Sadman Sadeed
Hu, Jianjun
contents Deep learning (DL) models have been widely used in materials property prediction with great success, especially for properties with large datasets. However, the out-of-distribution (OOD) performances of such models are questionable, especially when the training set is not large enough. Here we showed that using physical encoding rather than the widely used one-hot encoding can significantly improve the OOD performance by increasing the models' generalization performance, which is especially true for models trained with small datasets. Our benchmark results of both composition- and structure-based deep learning models over six datasets including formation energy, band gap, refractive index, and elastic properties predictions demonstrated the importance of physical encoding to OOD generalization for models trained on small datasets.
format Preprint
id arxiv_https___arxiv_org_abs_2407_15214
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Physical Encoding Improves OOD Performance in Deep Learning Materials Property Prediction
Fu, Nihang
Omee, Sadman Sadeed
Hu, Jianjun
Materials Science
Deep learning (DL) models have been widely used in materials property prediction with great success, especially for properties with large datasets. However, the out-of-distribution (OOD) performances of such models are questionable, especially when the training set is not large enough. Here we showed that using physical encoding rather than the widely used one-hot encoding can significantly improve the OOD performance by increasing the models' generalization performance, which is especially true for models trained with small datasets. Our benchmark results of both composition- and structure-based deep learning models over six datasets including formation energy, band gap, refractive index, and elastic properties predictions demonstrated the importance of physical encoding to OOD generalization for models trained on small datasets.
title Physical Encoding Improves OOD Performance in Deep Learning Materials Property Prediction
topic Materials Science
url https://arxiv.org/abs/2407.15214