Optimizing the architecture of Behler-Parrinello neural network potentials

Autor(en)

Lukáš Kývala, Christoph Dellago

Abstrakt

The architecture of neural network potentials is typically optimized at the beginning of the training process and remains unchanged throughout. Here, we investigate the accuracy of Behler-Parrinello neural network potentials for varying training set sizes. Using the QM9 and 3BPA datasets, we show that adjusting the network architecture according to the training set size improves the accuracy significantly. We demonstrate that both an insufficient and an excessive number of fitting parameters can have a detrimental impact on the accuracy of the neural network potential. Furthermore, we investigate the influences of descriptor complexity, neural network depth, and activation function on the model’s performance. We find that for the neural network potentials studied here, two hidden layers yield the best accuracy and that unbounded activation functions outperform bounded ones.

Organisation(en)

Computergestützte Physik und Physik der Weichen Materie

Journal

Journal of Chemical Physics

Band

159

Anzahl der Seiten

8

ISSN

0021-9606

DOI

https://doi.org/10.1063/5.0167260

Publikationsdatum

09-2023

Peer-reviewed

Ja

ÖFOS 2012

102019 Machine Learning, 103043 Computational Physics, 103029 Statistische Physik

ASJC Scopus Sachgebiete

Allgemeine Physik und Astronomie, Physical and Theoretical Chemistry

Link zum Portal

https://ucrisportal.univie.ac.at/de/publications/a14cd157-f5fc-48bb-9314-543b34703f41