notesum.ai

Published at November 27

What do physics-informed DeepONets learn? Understanding and improving training for scientific computing applications

cs.LG

cs.NA

math.NA

Released Date: November 27, 2024

Authors: Emily Williams¹, Amanda Howard¹, Brek Meuris², Panos Stinis¹

Aff.: ¹Advanced Computing, Mathematics and Data Division, Pacific Northwest National Laboratory, WA 99354; ²Sandia National Laboratories, NM 87123

Arxiv: http://arxiv.org/abs/2411.18459v1

Refer to caption

	$w$	Average relative $\ell_{2}$ error
data-driven	128	0.44% $\pm$ 0.32%
physics-informed (fixed)	128	0.48% $\pm$ 0.41%
physics-informed (NTK)	128	0.82% $\pm$ 0.54%