Disentangled Multi-Fidelity Deep Bayesian Active Learning

Dongxia Wu, Ruijia Niu, Matteo Chinazzi, Yian Ma, Rose Yu
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:37624-37634, 2023.

Abstract

To balance quality and cost, various domain areas of science and engineering run simulations at multiple levels of sophistication. Multi-fidelity active learning aims to learn a direct mapping from input parameters to simulation outputs at the highest fidelity by actively acquiring data from multiple fidelity levels. However, existing approaches based on Gaussian processes are hardly scalable to high-dimensional data. Deep learning-based methods often impose a hierarchical structure in hidden representations, which only supports passing information from low-fidelity to high-fidelity. These approaches can lead to the undesirable propagation of errors from low-fidelity representations to high-fidelity ones. We propose a novel framework called Disentangled Multi-fidelity Deep Bayesian Active Learning (D-MFDAL), which learns the surrogate models conditioned on the distribution of functions at multiple fidelities. On benchmark tasks of learning deep surrogates of partial differential equations including heat equation, Poisson’s equation and fluid simulations, our approach significantly outperforms state-of-the-art in prediction accuracy and sample efficiency.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-wu23p, title = {Disentangled Multi-Fidelity Deep {B}ayesian Active Learning}, author = {Wu, Dongxia and Niu, Ruijia and Chinazzi, Matteo and Ma, Yian and Yu, Rose}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {37624--37634}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/wu23p/wu23p.pdf}, url = {https://proceedings.mlr.press/v202/wu23p.html}, abstract = {To balance quality and cost, various domain areas of science and engineering run simulations at multiple levels of sophistication. Multi-fidelity active learning aims to learn a direct mapping from input parameters to simulation outputs at the highest fidelity by actively acquiring data from multiple fidelity levels. However, existing approaches based on Gaussian processes are hardly scalable to high-dimensional data. Deep learning-based methods often impose a hierarchical structure in hidden representations, which only supports passing information from low-fidelity to high-fidelity. These approaches can lead to the undesirable propagation of errors from low-fidelity representations to high-fidelity ones. We propose a novel framework called Disentangled Multi-fidelity Deep Bayesian Active Learning (D-MFDAL), which learns the surrogate models conditioned on the distribution of functions at multiple fidelities. On benchmark tasks of learning deep surrogates of partial differential equations including heat equation, Poisson’s equation and fluid simulations, our approach significantly outperforms state-of-the-art in prediction accuracy and sample efficiency.} }
Endnote
%0 Conference Paper %T Disentangled Multi-Fidelity Deep Bayesian Active Learning %A Dongxia Wu %A Ruijia Niu %A Matteo Chinazzi %A Yian Ma %A Rose Yu %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-wu23p %I PMLR %P 37624--37634 %U https://proceedings.mlr.press/v202/wu23p.html %V 202 %X To balance quality and cost, various domain areas of science and engineering run simulations at multiple levels of sophistication. Multi-fidelity active learning aims to learn a direct mapping from input parameters to simulation outputs at the highest fidelity by actively acquiring data from multiple fidelity levels. However, existing approaches based on Gaussian processes are hardly scalable to high-dimensional data. Deep learning-based methods often impose a hierarchical structure in hidden representations, which only supports passing information from low-fidelity to high-fidelity. These approaches can lead to the undesirable propagation of errors from low-fidelity representations to high-fidelity ones. We propose a novel framework called Disentangled Multi-fidelity Deep Bayesian Active Learning (D-MFDAL), which learns the surrogate models conditioned on the distribution of functions at multiple fidelities. On benchmark tasks of learning deep surrogates of partial differential equations including heat equation, Poisson’s equation and fluid simulations, our approach significantly outperforms state-of-the-art in prediction accuracy and sample efficiency.
APA
Wu, D., Niu, R., Chinazzi, M., Ma, Y. & Yu, R.. (2023). Disentangled Multi-Fidelity Deep Bayesian Active Learning. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:37624-37634 Available from https://proceedings.mlr.press/v202/wu23p.html.

Related Material