Abstract
Being able to quantify uncertainty when comparing a theoretical or computational model to observations is critical to conducting a sound scientific investigation. With the rise of data-driven modelling, understanding various sources of uncertainty and developing methods to estimate them has gained renewed attention. Five researchers discuss uncertainty quantification in machine-learned models with an emphasis on issues relevant to physics problems.
This is a preview of subscription content, access via your institution
Access options
Access Nature and 54 other Nature Portfolio journals
Get Nature+, our best-value online-access subscription
$29.99 / 30 days
cancel any time
Subscribe to this journal
Receive 12 digital issues and online access to articles
$99.00 per year
only $8.25 per issue
Buy this article
- Purchase on Springer Link
- Instant access to full article PDF
Prices may be subject to local taxes which are calculated during checkout
References
Cranmer, K., Brehmer, J. & Louppe, G. The frontier of simulation-based inference. Proc. Natl Acad. Sci. USA 117, 30055–30062 (2020).
Abdar, M. et al. A review of uncertainty quantification in deep learning: techniques, applications and challenges. Inf. Fusion 76, 243–297 (2021).
Hüllermeier, E. & Waegeman, W. Aleatoric and epistemic uncertainty in machine learning: an introduction to concepts and methods. Mach. Learn. 110, 457–506 (2021).
Gal, Y. & Ghahramani, Z. Dropout as a Bayesian approximation: representing model uncertainty in deep learning. In Proc. 33rd International Conference on International Conference on Machine Learning Vol 48 1050–1059 (PMLR, 2016).
Neal, R. M. Bayesian Learning for Neural Networks (Springer, 1996).
van Amersfoort, J., Smith, L., Teh, Y. W. & Gal, Y. Uncertainty estimation using a single deep deterministic neural network. In Proc. 37th International Conference on Machine Learning Vol 119 9690–9700 (PMLR, 2020).
Liu, J. Z. et al. Simple and principled uncertainty estimation with deterministic deep learning via distance awareness. In Proc. 34th International Conference on Neural Information Processing Systems (NIPS’20) 7498–7512 (Curran Associates, 2020).
van Amersfoort, J., Smith, L., Jesson, A., Key, O. & Gal, Y. Improving deterministic uncertainty estimation in deep learning for classification and regression. Preprint at https://arxiv.org/abs/2102.11409v1 (2021).
Martin, S. M. et al. Korali: efficient and scalable software framework for Bayesian uncertainty quantification and stochastic optimization. Comput. Methods Appl. Mech. Eng. 389, 114264 (2022).
Bae, H. J. & Koumoutsakos, P. Scientific multi-agent reinforcement learning for wall-models of turbulent flows. Nat. Commun. 13, 1443 (2022).
Berger, J. O. & Smith, L. A. On the statistical formalism of uncertainty quantification. Annu. Rev. Stat. Appl. 6, 433–460 (2019).
Acknowledgements
Y.G. holds a Turing Articifical Intelligence Fellowship at the Alan Turing Institute, which is supported by Engineering and Physical Sciences Research Council (EPSRC) grant reference V030302/1.
Author information
Authors and Affiliations
Contributions
Yarin Gal is Associate Professor of Machine Learning at the University of Oxford Computer Science department, UK, and leads the Oxford Applied and Theoretical Machine Learning (OATML) group. He has made substantial contributions to early work in modern Bayesian deep learning — quantifying uncertainty in deep learning — and developed machine learning and artificial intelligence tools that can inform their users when the tools are ‘guessing at random’. These tools have been deployed widely in industry and academia, such as in medical applications, robotics, computer vision, astronomy and sciences, and are also used by NASA.
Petros Koumoutsakos is the Herbert S. Winokur Professor of Science and Engineering in the John A. Paulson School of Engineering and Applied Sciences at Harvard University. He is elected Fellow of the American Society of Mechanical Engineers (ASME), the American Physical Society (APS), the Society of Industrial and Applied Mathematics (SIAM) and a recipient of the SIAM/ACM Gordon Bell award in Supercomputing. He is an international member of the US National Academy of Engineering (NAE). Petros’ research interests are on the fundamentals and applications of computing and artificial intelligence to understand, predict and optimize complex systems in engineering, nanotechnology and medicine.
Francois Lanusse is a CNRS researcher, part of CosmoStat Laboratory at CEA Saclay. Previously, he worked at the Berkeley Center for Cosmological Physics, the Foundation of Data Analysis Institute at UC Berkeley and in the McWilliams Center for Cosmology at Carnegie Mellon University. His research is at the intersection of cosmology and machine learning.
Gilles Louppe is Associate Professor in Artificial Intelligence and Deep Learning at the University of Liège in Belgium. His research is at the intersection of deep learning, approximate inference and the physical sciences. He has been developing a new generation of simulation-based inference algorithms based on deep learning, with several applications in particle physics, astrophysics, astronomy and gravitational wave science.
Costas Papadimitriou is Professor of Structural Dynamics and the founding Director of System Dynamics Laboratory at the University of Thessaly in Greece. His research interests include data-driven uncertainty quantification in engineering and applied sciences, optimal experimental design, structural health monitoring and reliability. He is the recipient of the 2014 European Association of Structural Dynamics (EASD) Senior Award in Computational Structural Dynamics.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Author contributions
All authors contributed equally to the preparation of this manuscript.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Gal, Y., Koumoutsakos, P., Lanusse, F. et al. Bayesian uncertainty quantification for machine-learned models in physics. Nat Rev Phys 4, 573–577 (2022). https://doi.org/10.1038/s42254-022-00498-4
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/s42254-022-00498-4
