{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,3,7]],"date-time":"2026-03-07T18:24:48Z","timestamp":1772907888877,"version":"3.50.1"},"reference-count":46,"publisher":"IOP Publishing","issue":"1","license":[{"start":{"date-parts":[[2023,2,6]],"date-time":"2023-02-06T00:00:00Z","timestamp":1675641600000},"content-version":"vor","delay-in-days":0,"URL":"http:\/\/creativecommons.org\/licenses\/by\/4.0"},{"start":{"date-parts":[[2023,2,6]],"date-time":"2023-02-06T00:00:00Z","timestamp":1675641600000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/iopscience.iop.org\/info\/page\/text-and-data-mining"}],"funder":[{"DOI":"10.13039\/100000183","name":"Army Research Office","doi-asserted-by":"crossref","award":["W911NF-19-2-0333"],"award-info":[{"award-number":["W911NF-19-2-0333"]}],"id":[{"id":"10.13039\/100000183","id-type":"DOI","asserted-by":"crossref"}]},{"DOI":"10.13039\/100000002","name":"National Institutes of Health","doi-asserted-by":"crossref","award":["NIH-R01GM117594"],"award-info":[{"award-number":["NIH-R01GM117594"]}],"id":[{"id":"10.13039\/100000002","id-type":"DOI","asserted-by":"crossref"}]},{"DOI":"10.13039\/100000015","name":"U.S. Department of Energy","doi-asserted-by":"crossref","award":["DE-SC0007890"],"award-info":[{"award-number":["DE-SC0007890"]}],"id":[{"id":"10.13039\/100000015","id-type":"DOI","asserted-by":"crossref"}]}],"content-domain":{"domain":["iopscience.iop.org"],"crossmark-restriction":false},"short-container-title":["Mach. Learn.: Sci. Technol."],"published-print":{"date-parts":[[2023,3,1]]},"abstract":"<jats:title>Abstract<\/jats:title>\n               <jats:p>Physics-informed neural networks (PINNs) have been shown to be effective in solving partial differential equations by capturing the physics induced constraints as a part of the training loss function. This paper shows that a PINN can be sensitive to errors in training data and overfit itself in dynamically propagating these errors over the domain of the solution of the PDE. It also shows how physical regularizations based on continuity criteria and conservation laws fail to address this issue and rather introduce problems of their own causing the deep network to converge to a physics-obeying local minimum instead of the global minimum. We introduce Gaussian process (GP) based smoothing that recovers the performance of a PINN and promises a robust architecture against noise\/errors in measurements. Additionally, we illustrate an inexpensive method of quantifying the evolution of uncertainty based on the variance estimation of GPs on boundary data. Robust PINN performance is also shown to be achievable by choice of sparse sets of inducing points based on sparsely induced GPs. We demonstrate the performance of our proposed methods and compare the results from existing benchmark models in literature for time-dependent Schr\u00f6dinger and Burgers\u2019 equations.<\/jats:p>","DOI":"10.1088\/2632-2153\/acb416","type":"journal-article","created":{"date-parts":[[2023,1,17]],"date-time":"2023-01-17T22:52:49Z","timestamp":1673995969000},"page":"015013","update-policy":"https:\/\/doi.org\/10.1088\/crossmark-policy","source":"Crossref","is-referenced-by-count":27,"title":["Recipes for when physics fails: recovering robust learning of physics informed neural networks"],"prefix":"10.1088","volume":"4","author":[{"given":"Chandrajit","family":"Bajaj","sequence":"first","affiliation":[]},{"given":"Luke","family":"McLennan","sequence":"additional","affiliation":[]},{"given":"Timothy","family":"Andeen","sequence":"additional","affiliation":[]},{"ORCID":"https:\/\/orcid.org\/0000-0002-0116-1012","authenticated-orcid":true,"given":"Avik","family":"Roy","sequence":"additional","affiliation":[]}],"member":"266","published-online":{"date-parts":[[2023,2,6]]},"reference":[{"key":"mlstacb416bib1","doi-asserted-by":"publisher","first-page":"361","DOI":"10.1146\/annurev-physchem-042018-052331","article-title":"Machine learning for molecular simulation","volume":"71","author":"No\u00e9","year":"2020","journal-title":"Annu. Rev. Phys. Chem."},{"key":"mlstacb416bib2","doi-asserted-by":"publisher","first-page":"115","DOI":"10.1142\/S0129065709001884","article-title":"Nonlinear system identification based on internal recurrent neural networks","volume":"19","author":"Puscasu","year":"2009","journal-title":"Int. J. Neural Syst."},{"key":"mlstacb416bib3","doi-asserted-by":"publisher","first-page":"5322","DOI":"10.1038\/s41467-018-07520-3","article-title":"Constructing exact representations of quantum many-body systems with deep neural networks","volume":"9","author":"Carleo","year":"2018","journal-title":"Nat. Commun."},{"key":"mlstacb416bib4","doi-asserted-by":"publisher","first-page":"418","DOI":"10.1006\/jcph.2002.7176","article-title":"Preconditioning techniques for large linear systems: a survey","volume":"182","author":"Benzi","year":"2002","journal-title":"J. Comput. Phys."},{"key":"mlstacb416bib5","volume":"vol 154","author":"Grossmann","year":"2007"},{"key":"mlstacb416bib6","doi-asserted-by":"publisher","first-page":"195","DOI":"10.1002\/cnm.1640100303","article-title":"Neural-network-based approximations for solving partial differential equations","volume":"10","author":"Dissanayake","year":"1994","journal-title":"Commun. Numer. Methods Eng."},{"key":"mlstacb416bib7","doi-asserted-by":"publisher","first-page":"261","DOI":"10.1023\/A:1012784129883","article-title":"Neural network method for solving partial differential equations","volume":"14","author":"Aarts","year":"2001","journal-title":"Neural Process. Lett."},{"key":"mlstacb416bib8","doi-asserted-by":"publisher","first-page":"2812","DOI":"10.3923\/jas.2007.2812.2817","article-title":"Feedforward neural network for solving partial differential equations","volume":"7","author":"Hayati","year":"2007","journal-title":"J. Appl. Sci."},{"key":"mlstacb416bib9","doi-asserted-by":"publisher","first-page":"5595","DOI":"10.5555\/3122009.3242010","article-title":"Automatic differentiation in machine learning: a survey","volume":"18","author":"Baydin","year":"2018","journal-title":"J. Mach. Learn. Res."},{"key":"mlstacb416bib10","doi-asserted-by":"publisher","first-page":"987","DOI":"10.1109\/72.712178","article-title":"Artificial neural networks for solving ordinary and partial differential equations","volume":"9","author":"Lagaris","year":"1998","journal-title":"IEEE Trans. Neural Netw."},{"key":"mlstacb416bib11","article-title":"Neural ordinary differential equations","author":"Ricky","year":"2018"},{"key":"mlstacb416bib12","author":"Pontryagin","year":"1987"},{"key":"mlstacb416bib13","article-title":"Ode2vae: Deep generative second order odes with bayesian neural networks","author":"Y\u0131ld\u0131z","year":"2019"},{"key":"mlstacb416bib14","doi-asserted-by":"publisher","first-page":"686","DOI":"10.1016\/j.jcp.2018.10.045","article-title":"Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations","volume":"378","author":"Raissi","year":"2019","journal-title":"J. Comput. Phys."},{"key":"mlstacb416bib15","doi-asserted-by":"publisher","first-page":"359","DOI":"10.1016\/0893-6080(89)90020-8","article-title":"Multilayer feedforward networks are universal approximators","volume":"2","author":"Hornik","year":"1989","journal-title":"Neural Netw."},{"key":"mlstacb416bib16","doi-asserted-by":"publisher","first-page":"143","DOI":"10.1017\/S0962492900002919","article-title":"Approximation theory of the mlp model in neural networks","volume":"8","author":"Pinkus","year":"1999","journal-title":"Acta Numer."},{"key":"mlstacb416bib17","doi-asserted-by":"publisher","first-page":"A2603","DOI":"10.1137\/18M1229845","article-title":"fpinns: Fractional physics-informed neural networks","volume":"41","author":"Pang","year":"2019","journal-title":"SIAM J. Sci. Comput."},{"key":"mlstacb416bib18","article-title":"B-pinns: Bayesian physics-informed neural networks for forward and inverse pde problems with noisy data","author":"Yang","year":"2020"},{"key":"mlstacb416bib19","doi-asserted-by":"publisher","DOI":"10.1016\/j.jcp.2019.109136","article-title":"Adaptive activation functions accelerate convergence in deep and physics-informed neural networks","volume":"404","author":"Jagtap","year":"2020","journal-title":"J. Comput. Phys."},{"key":"mlstacb416bib20","doi-asserted-by":"publisher","first-page":"136","DOI":"10.1016\/j.jcp.2019.05.027","article-title":"Adversarial uncertainty quantification in physics-informed neural networks","volume":"394","author":"Yang","year":"2019","journal-title":"J. Comput. Phys."},{"key":"mlstacb416bib21","doi-asserted-by":"publisher","DOI":"10.1016\/j.cma.2020.113250","article-title":"Ppinn: parareal physics-informed neural network for time-dependent pdes","volume":"370","author":"Meng","year":"2020","journal-title":"Comput. Methods Appl. Mech. Eng."},{"key":"mlstacb416bib22","doi-asserted-by":"publisher","DOI":"10.1016\/j.cma.2020.113028","article-title":"Conservative physics-informed neural networks on discrete domains for conservation laws: applications to forward and inverse problems","volume":"365","author":"Jagtap","year":"2020","journal-title":"Comput. Methods Appl. Mech. Eng."},{"key":"mlstacb416bib23","doi-asserted-by":"publisher","first-page":"2042","DOI":"10.4208\/cicp.OA-2020-0193","article-title":"On the convergence of physics informed neural networks for linear second-order elliptic and parabolic type pdes","volume":"28","author":"Shin","year":"2020","journal-title":"Commun. Comput. Phys."},{"key":"mlstacb416bib24","article-title":"Variational physics-informed neural networks for solving partial differential equations","author":"Kharazmi","year":"2019"},{"key":"mlstacb416bib25","article-title":"The derivation of conservation laws for nonlinear schrodinger equations with a power type nonlinearity","author":"Fujiwara","year":"2014"},{"key":"mlstacb416bib26","doi-asserted-by":"publisher","first-page":"1653","DOI":"10.1143\/JPSJ.46.1653","article-title":"Method of conservation laws for solving nonlinear schrodinger equation","volume":"46","author":"Watanabe","year":"1978","journal-title":"J. Phys. Soc. Japan"},{"key":"mlstacb416bib27","article-title":"The Local Conservation Laws of the Nonlinear Schrodinger Equation","author":"Barrett","year":"2013"},{"key":"mlstacb416bib28","doi-asserted-by":"publisher","first-page":"225","DOI":"10.1090\/qam\/42889","article-title":"On a quasi-linear parabolic equation occurring in aerodynamics","volume":"9","author":"Cole","year":"1951","journal-title":"Q. Appl. Math."},{"key":"mlstacb416bib29","doi-asserted-by":"publisher","first-page":"201","DOI":"10.1002\/cpa.3160030302","article-title":"The partial differential equation ut+uux=\u03bcuxx","volume":"3","author":"Hopf","year":"1950","journal-title":"Commun. Pure Appl. Math."},{"key":"mlstacb416bib30","volume":"vol 2","author":"Williams","year":"2006"},{"key":"mlstacb416bib31","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1016\/j.jmp.2018.03.001","article-title":"A tutorial on gaussian process regression: modelling, exploring and exploiting functions","volume":"85","author":"Schulz","year":"2018","journal-title":"J. Math. Psychol."},{"key":"mlstacb416bib32","author":"Gramacy","year":"2020"},{"key":"mlstacb416bib33","first-page":"pp 7759","article-title":"Robustness guarantees for bayesian inference with gaussian processes","volume":"vol 33","author":"Cardelli","year":"2019"},{"key":"mlstacb416bib34","first-page":"pp 3372","article-title":"Adversarial robustness guarantees for classification with gaussian processes","author":"Blaas","year":"2020"},{"key":"mlstacb416bib35","article-title":"Additive gaussian processes","volume":"vol 24","author":"Duvenaud","year":"2011"},{"key":"mlstacb416bib36","first-page":"pp 1166","article-title":"Structure discovery in nonparametric regression through compositional kernel search","author":"Duvenaud","year":"2013"},{"key":"mlstacb416bib37","doi-asserted-by":"publisher","DOI":"10.1103\/PhysRevLett.121.255702","article-title":"Extrapolating quantum observables with machine learning: inferring multiple phase transitions from properties of a single phase","volume":"121","author":"Vargas-Hern\u00e1ndez","year":"2018","journal-title":"Phys. Rev. Lett."},{"key":"mlstacb416bib38","first-page":"pp 234","article-title":"Solving noisy linear operator equations by gaussian processes: application to ordinary and partial differential equations","volume":"vol 3","author":"Graepel","year":"2003"},{"key":"mlstacb416bib39","doi-asserted-by":"publisher","first-page":"A172","DOI":"10.1137\/17M1120762","article-title":"Numerical gaussian processes for time-dependent and nonlinear partial differential equations","volume":"40","author":"Raissi","year":"2018","journal-title":"SIAM J. Sci. Comput."},{"key":"mlstacb416bib40","doi-asserted-by":"crossref","DOI":"10.1016\/j.jcp.2021.110668","article-title":"Solving and learning nonlinear pdes with gaussian processes","author":"Chen","year":"2021"},{"key":"mlstacb416bib41","article-title":"Sample efficient reinforcement learning with gaussian processes","author":"Grande","year":"2014"},{"key":"mlstacb416bib42","article-title":"Sample-efficient reinforcement learning using deep gaussian processes","author":"Gadd","year":"2020"},{"key":"mlstacb416bib43","article-title":"The variational gaussian process","author":"Tran","year":"2016"},{"key":"mlstacb416bib44","article-title":"Adaptive inducing points selection for gaussian processes","author":"Galy-Fajou","year":"2020"},{"key":"mlstacb416bib45","doi-asserted-by":"publisher","first-page":"213","DOI":"10.1002\/fld.1650030302","article-title":"Generating exact solutions of the two-dimensional burgers\u2019 equations","volume":"3","author":"Fletcher","year":"1983","journal-title":"Int. J. Numer. Methods Fluids"},{"key":"mlstacb416bib46","doi-asserted-by":"publisher","first-page":"840","DOI":"10.1016\/j.camwa.2010.05.031","article-title":"Numerical solutions of two-dimensional burgers\u2019 equations by discrete adomian decomposition method","volume":"60","author":"Zhu","year":"2010","journal-title":"Comput. Math. Appl."}],"container-title":["Machine Learning: Science and Technology"],"original-title":[],"link":[{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/acb416","content-type":"text\/html","content-version":"am","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/acb416\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/acb416","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/acb416\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/acb416\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"syndication"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/acb416\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/acb416\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"similarity-checking"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/acb416\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2023,2,6]],"date-time":"2023-02-06T11:17:54Z","timestamp":1675682274000},"score":1,"resource":{"primary":{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/acb416"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2023,2,6]]},"references-count":46,"journal-issue":{"issue":"1","published-online":{"date-parts":[[2023,2,6]]},"published-print":{"date-parts":[[2023,3,1]]}},"URL":"https:\/\/doi.org\/10.1088\/2632-2153\/acb416","relation":{},"ISSN":["2632-2153"],"issn-type":[{"value":"2632-2153","type":"electronic"}],"subject":[],"published":{"date-parts":[[2023,2,6]]},"assertion":[{"value":"Recipes for when physics fails: recovering robust learning of physics informed neural networks","name":"article_title","label":"Article Title"},{"value":"Machine Learning: Science and Technology","name":"journal_title","label":"Journal Title"},{"value":"paper","name":"article_type","label":"Article Type"},{"value":"\u00a9 2023 The Author(s). Published by IOP Publishing Ltd","name":"copyright_information","label":"Copyright Information"},{"value":"2022-07-26","name":"date_received","label":"Date Received","group":{"name":"publication_dates","label":"Publication dates"}},{"value":"2023-01-17","name":"date_accepted","label":"Date Accepted","group":{"name":"publication_dates","label":"Publication dates"}},{"value":"2023-02-06","name":"date_epub","label":"Online publication date","group":{"name":"publication_dates","label":"Publication dates"}}]}}