Equation Discovery with Bayesian Spike-and-Slab Priors and Efficient Kernels

Da Long, Wei Xing, Aditi Krishnapriyan, Robert Kirby, Shandian Zhe, Michael W. Mahoney
Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, PMLR 238:2413-2421, 2024.

Abstract

Discovering governing equations from data is important to many scientific and engineering applications. Despite promising successes, existing methods are still challenged by data sparsity and noise issues, both of which are ubiquitous in practice. Moreover, state-of-the-art methods lack uncertainty quantification and/or are costly in training. To overcome these limitations, we propose a novel equation discovery method based on Kernel learning and BAyesian Spike-and-Slab priors (KBASS). We use kernel regression to estimate the target function, which is flexible, expressive, and more robust to data sparsity and noises. We combine it with a Bayesian spike-and-slab prior — an ideal Bayesian sparse distribution — for effective operator selection and uncertainty quantification. We develop an expectation-propagation expectation-maximization (EP-EM) algorithm for efficient posterior inference and function estimation. To overcome the computational challenge of kernel regression, we place the function values on a mesh and induce a Kronecker product construction, and we use tensor algebra to enable efficient computation and optimization. We show the advantages of KBASS on a list of benchmark ODE and PDE discovery tasks. The code is available at \url{https://github.com/long-da/KBASS}.

Cite this Paper


BibTeX
@InProceedings{pmlr-v238-long24a, title = {Equation Discovery with {B}ayesian Spike-and-Slab Priors and Efficient Kernels}, author = {Long, Da and Xing, Wei and Krishnapriyan, Aditi and Kirby, Robert and Zhe, Shandian and W. Mahoney, Michael}, booktitle = {Proceedings of The 27th International Conference on Artificial Intelligence and Statistics}, pages = {2413--2421}, year = {2024}, editor = {Dasgupta, Sanjoy and Mandt, Stephan and Li, Yingzhen}, volume = {238}, series = {Proceedings of Machine Learning Research}, month = {02--04 May}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v238/long24a/long24a.pdf}, url = {https://proceedings.mlr.press/v238/long24a.html}, abstract = {Discovering governing equations from data is important to many scientific and engineering applications. Despite promising successes, existing methods are still challenged by data sparsity and noise issues, both of which are ubiquitous in practice. Moreover, state-of-the-art methods lack uncertainty quantification and/or are costly in training. To overcome these limitations, we propose a novel equation discovery method based on Kernel learning and BAyesian Spike-and-Slab priors (KBASS). We use kernel regression to estimate the target function, which is flexible, expressive, and more robust to data sparsity and noises. We combine it with a Bayesian spike-and-slab prior — an ideal Bayesian sparse distribution — for effective operator selection and uncertainty quantification. We develop an expectation-propagation expectation-maximization (EP-EM) algorithm for efficient posterior inference and function estimation. To overcome the computational challenge of kernel regression, we place the function values on a mesh and induce a Kronecker product construction, and we use tensor algebra to enable efficient computation and optimization. We show the advantages of KBASS on a list of benchmark ODE and PDE discovery tasks. The code is available at \url{https://github.com/long-da/KBASS}.} }
Endnote
%0 Conference Paper %T Equation Discovery with Bayesian Spike-and-Slab Priors and Efficient Kernels %A Da Long %A Wei Xing %A Aditi Krishnapriyan %A Robert Kirby %A Shandian Zhe %A Michael W. Mahoney %B Proceedings of The 27th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2024 %E Sanjoy Dasgupta %E Stephan Mandt %E Yingzhen Li %F pmlr-v238-long24a %I PMLR %P 2413--2421 %U https://proceedings.mlr.press/v238/long24a.html %V 238 %X Discovering governing equations from data is important to many scientific and engineering applications. Despite promising successes, existing methods are still challenged by data sparsity and noise issues, both of which are ubiquitous in practice. Moreover, state-of-the-art methods lack uncertainty quantification and/or are costly in training. To overcome these limitations, we propose a novel equation discovery method based on Kernel learning and BAyesian Spike-and-Slab priors (KBASS). We use kernel regression to estimate the target function, which is flexible, expressive, and more robust to data sparsity and noises. We combine it with a Bayesian spike-and-slab prior — an ideal Bayesian sparse distribution — for effective operator selection and uncertainty quantification. We develop an expectation-propagation expectation-maximization (EP-EM) algorithm for efficient posterior inference and function estimation. To overcome the computational challenge of kernel regression, we place the function values on a mesh and induce a Kronecker product construction, and we use tensor algebra to enable efficient computation and optimization. We show the advantages of KBASS on a list of benchmark ODE and PDE discovery tasks. The code is available at \url{https://github.com/long-da/KBASS}.
APA
Long, D., Xing, W., Krishnapriyan, A., Kirby, R., Zhe, S. & W. Mahoney, M.. (2024). Equation Discovery with Bayesian Spike-and-Slab Priors and Efficient Kernels. Proceedings of The 27th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 238:2413-2421 Available from https://proceedings.mlr.press/v238/long24a.html.

Related Material