Details

Autor(en) / Beteiligte

• Kasim, M F
• Watson-Parris, D
• Deaconu, L
• Oliver, S
• Hatfield, P
• Froula, D H
• Gregori, G
• Jarvis, M
• Khatiwala, S
• Korenaga, J
• Topp-Mugglestone, J
• Viezzer, E
• Vinko, S M

Titel

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

Ist Teil von

• Machine learning: science and technology, 2022-03, Vol.3 (1), p.15013

Ort / Verlag

Bristol: IOP Publishing

Erscheinungsjahr

2022

Quelle

Free E-Journal (出版社公開部分のみ）

Beschreibungen/Notizen

• 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.