Building high accuracy emulators for scientific simulations with deep neural architecture search

Kasim, M F and Watson-Parris, D and Deaconu, L and Oliver, S and Hatfield, P and Froula, D H and Gregori, G and Jarvis, M and Khatiwala, S and Korenaga, J and Topp-Mugglestone, J and Viezzer, E and Vinko, S M (2022) Building high accuracy emulators for scientific simulations with deep neural architecture search. Machine Learning: Science and Technology, 3 (1). 015013. ISSN 2632-2153

[thumbnail of Kasim_2022_Mach._Learn.__Sci._Technol._3_015013.pdf] Text
Kasim_2022_Mach._Learn.__Sci._Technol._3_015013.pdf - Published Version

Download (2MB)

Abstract

Computer simulations are invaluable tools for scientific discovery. However, accurate simulations are often slow to execute, which limits their applicability to extensive parameter exploration, large-scale data analysis, and uncertainty quantification. A promising route to accelerate simulations by building fast emulators with machine learning requires large training datasets, which can be prohibitively expensive to obtain with slow simulations. Here we present a method based on neural architecture search to build accurate emulators even with a limited number of training data. The method successfully emulates simulations in 10 scientific cases including astrophysics, climate science, biogeochemistry, high energy density physics, fusion energy, and seismology, using the same super-architecture, algorithm, and hyperparameters. Our approach also inherently provides emulator uncertainty estimation, adding further confidence in their use. We anticipate this work will accelerate research involving expensive simulations, allow more extensive parameters exploration, and enable new, previously unfeasible computational discovery.

Item Type: Article
Subjects: Open Archive Press > Multidisciplinary
Depositing User: Unnamed user with email support@openarchivepress.com
Date Deposited: 10 Jul 2023 05:06
Last Modified: 11 May 2024 09:40
URI: http://library.2pressrelease.co.in/id/eprint/1714

Actions (login required)

View Item
View Item