{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,8,17]],"date-time":"2026-08-17T23:10:33Z","timestamp":1787008233257,"version":"build-2736575974"},"reference-count":37,"publisher":"IOP Publishing","issue":"4","license":[{"start":{"date-parts":[[2025,10,10]],"date-time":"2025-10-10T00:00:00Z","timestamp":1760054400000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"},{"start":{"date-parts":[[2025,10,10]],"date-time":"2025-10-10T00:00:00Z","timestamp":1760054400000},"content-version":"tdm","delay-in-days":0,"URL":"https:\/\/iopscience.iop.org\/info\/page\/text-and-data-mining"}],"funder":[{"DOI":"10.13039\/501100003185","name":"Fraunhofer-Gesellschaft","doi-asserted-by":"crossref","award":["PREPARE 40-08394"],"award-info":[{"award-number":["PREPARE 40-08394"]}],"id":[{"id":"10.13039\/501100003185","id-type":"DOI","asserted-by":"crossref"}]},{"DOI":"10.13039\/501100001659","name":"Deutsche Forschungsgemeinschaft","doi-asserted-by":"crossref","award":["CRC 1639 NuMeriQS - project no. 511713970"],"award-info":[{"award-number":["CRC 1639 NuMeriQS - project no. 511713970"]}],"id":[{"id":"10.13039\/501100001659","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":[[2025,12,30]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>We present an approach for end-to-end training of machine learning models for structure-property modeling on collections of datasets derived using different density functional theory functionals and basis sets. This approach overcomes the problem of data inconsistencies in the training of machine learning models on atomistic data. We rephrase the underlying problem as a multi-task learning scenario. We show that conditioning neural network-based models on trainable embedding vectors can effectively account for quantitative differences between methods. This allows for joint training on multiple datasets that would otherwise be incompatible. Therefore, this procedure circumvents the need for re-computations at a unified level of theory. Numerical experiments demonstrate that training on multiple reference methods enables transfer learning between tasks, resulting in even lower errors compared to training on separate tasks alone. Furthermore, we show that this approach can be used for multi-fidelity learning, improving data efficiency for the highest fidelity by an order of magnitude. To test scalability, we train a single model on a joint dataset compiled from ten disjoint subsets of the MultiXC-QM9 dataset generated by different reference methods. Again, we observe transfer learning effects that improve the model errors by a factor of 2 compared to training on each subset alone. We extend our investigation to machine learning force fields for material simulations. To this end, we incorporate trainable embedding vectors into the readout layer of a deep graph neural network (M3GNet) that is simultaneously trained on PBE and r2SCAN labels of the MatPES dataset. We observe that joint training on both fidelity levels reduces the amount of r2SCAN data required to achieve the accuracy of a single-fidelity model by a factor of 10.<\/jats:p>","DOI":"10.1088\/2632-2153\/ae0d41","type":"journal-article","created":{"date-parts":[[2025,9,29]],"date-time":"2025-09-29T22:55:52Z","timestamp":1759186552000},"page":"045004","update-policy":"https:\/\/doi.org\/10.1088\/crossmark-policy","source":"Crossref","is-referenced-by-count":1,"title":["Multi-fidelity learning for atomistic models via trainable data embeddings"],"prefix":"10.1088","volume":"6","author":[{"ORCID":"https:\/\/orcid.org\/0009-0001-1797-5686","authenticated-orcid":true,"given":"Rick","family":"Oerder","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0009-0000-4807-9609","authenticated-orcid":true,"given":"Gerrit","family":"Schmieden","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-5399-043X","authenticated-orcid":true,"given":"Jan","family":"Hamaekers","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"266","published-online":{"date-parts":[[2025,10,10]]},"reference":[{"key":"mlstae0d41bib1","article-title":"A foundation model for atomistic materials chemistry","author":"Batatia","year":"2023","type":"preprint"},{"key":"mlstae0d41bib2","doi-asserted-by":"publisher","first-page":"1032","DOI":"10.1002\/qua.24890","type":"journal-article","article-title":"Constructing high-dimensional neural network potentials: a tutorial review","volume":"115","author":"Behler","year":"2015","journal-title":"Int. J. Quantum Chem."},{"key":"mlstae0d41bib3","doi-asserted-by":"publisher","first-page":"08","DOI":"10.1038\/sdata.2014.22","type":"journal-article","article-title":"Quantum chemistry structures and properties of 134 kilo molecules","volume":"1","author":"Ramakrishnan","year":"2014","journal-title":"Sci. Data"},{"key":"mlstae0d41bib4","doi-asserted-by":"publisher","first-page":"185","DOI":"10.1038\/s41597-022-01288-4","type":"journal-article","article-title":"Geom, energy-annotated molecular conformations for property prediction and molecular generation","volume":"9","author":"Axelrod","year":"2022","journal-title":"Sci. Data"},{"key":"mlstae0d41bib5","doi-asserted-by":"publisher","first-page":"01","DOI":"10.1038\/s41597-022-01882-6","type":"journal-article","article-title":"Spice, a dataset of drug-like molecules and peptides for training machine learning potentials","volume":"10","author":"Eastman","year":"2023","journal-title":"Sci. Data"},{"key":"mlstae0d41bib6","doi-asserted-by":"publisher","first-page":"02","DOI":"10.1038\/s41597-021-00812-2","type":"journal-article","article-title":"QM7-X, a comprehensive dataset of quantum-mechanical properties spanning the chemical space of small organic molecules","volume":"8","author":"Hoja","year":"2021","journal-title":"Sci. Data"},{"key":"mlstae0d41bib7","doi-asserted-by":"publisher","DOI":"10.1088\/2632-2153\/ad8f13","type":"journal-article","article-title":"Molecular quantum chemical data sets and databases for machine learning potentials","volume":"5","author":"Ullah","year":"2024","journal-title":"Mach. Learn.: Sci. Technol."},{"key":"mlstae0d41bib8","doi-asserted-by":"publisher","first-page":"nwae403","DOI":"10.1093\/nsr\/nwae403","type":"journal-article","article-title":"A survey on multimodal large language models","volume":"11","author":"Yin","year":"2024","journal-title":"Natl Sci. Rev."},{"key":"mlstae0d41bib9","doi-asserted-by":"publisher","DOI":"10.1038\/s41467-023-37477-x","type":"journal-article","article-title":"Explainable multi-task learning for multi-modality biological data analysis","volume":"14 2546","author":"Tang","year":"2023","journal-title":"Nat. Commun."},{"key":"mlstae0d41bib10","doi-asserted-by":"publisher","first-page":"5383","DOI":"10.1039\/D2CP05793J","type":"journal-article","article-title":"Transfer learning for chemically accurate interatomic neural network potentials","volume":"25","author":"Zaverkin","year":"2023","journal-title":"Phys. Chem. Chem. Phys."},{"key":"mlstae0d41bib11","doi-asserted-by":"publisher","first-page":"120","DOI":"10.1039\/D4FD00107A","type":"journal-article","article-title":"Data-efficient fine-tuning of foundational models for first-principles quality sublimation enthalpies","volume":"256","author":"Kaur","year":"2025","journal-title":"Faraday Discuss."},{"key":"mlstae0d41bib12","doi-asserted-by":"publisher","DOI":"10.1088\/2632-2153\/ad2cef","type":"journal-article","article-title":"Optimized multifidelity machine learning for quantum chemistry","volume":"5","author":"Vinod","year":"2024","journal-title":"Mach. Learn.: Sci. Technol."},{"key":"mlstae0d41bib13","doi-asserted-by":"publisher","DOI":"10.1002\/jcc.70056","type":"journal-article","article-title":"Predicting molecular energies of small organic molecules with multi-fidelity methods","volume":"46","author":"Vinod","year":"2025","journal-title":"J. Comput. Chem."},{"key":"mlstae0d41bib14","article-title":"On first-order meta-learning algorithms","author":"Nichol","year":"2018","type":"preprint"},{"key":"mlstae0d41bib15","doi-asserted-by":"publisher","first-page":"07","DOI":"10.1038\/s41524-024-01339-x","type":"journal-article","article-title":"Learning together: Towards foundation models for machine learning interatomic potentials with meta-learning","volume":"10","author":"Allen","year":"2024","journal-title":"npj Comput. Mater."},{"key":"mlstae0d41bib16","doi-asserted-by":"publisher","DOI":"10.1063\/5.0201681","type":"journal-article","article-title":"Multitask methods for predicting molecular properties from heterogeneous data","volume":"161","author":"Fisher","year":"2024","journal-title":"J. Chem. Phys."},{"key":"mlstae0d41bib17","doi-asserted-by":"crossref","DOI":"10.26434\/chemrxiv-2024-ng3ws","type":"preprint","article-title":"All-in-one foundational models learning across quantum chemical levels","author":"Chen","year":"2024"},{"key":"mlstae0d41bib18","article-title":"Data-efficient multi-fidelity training for high-fidelity machine learning interatomic potentials","author":"Kim","year":"2024","type":"preprint"},{"key":"mlstae0d41bib19","doi-asserted-by":"publisher","first-page":"46","DOI":"10.1038\/s43588-020-00002-x","type":"journal-article","article-title":"Learning properties of ordered and disordered materials from multi-fidelity data","volume":"1","author":"Chen","year":"2021","journal-title":"Nat. Computat. Sci."},{"key":"mlstae0d41bib20","doi-asserted-by":"publisher","first-page":"65","DOI":"10.1038\/s41524-025-01550-4","type":"journal-article","article-title":"Data-efficient construction of high-fidelity graph deep learning interatomic potentials","volume":"11","author":"Ko","year":"2025","journal-title":"npj Comput. Mater."},{"key":"mlstae0d41bib21","doi-asserted-by":"publisher","first-page":"783","DOI":"10.1038\/s41597-023-02690-2","type":"journal-article","article-title":"MultiXC-QM9: large dataset of molecular and reaction energies from multi-level quantum chemical methods","volume":"10","author":"Nandi","year":"2023","journal-title":"Sci. Data"},{"key":"mlstae0d41bib22","doi-asserted-by":"publisher","first-page":"13681","DOI":"10.1021\/acs.chemrev.4c00572","type":"journal-article","article-title":"Data generation for machine learning interatomic potentials and beyond","volume":"124","author":"Kulichenko","year":"2024","journal-title":"Chem. Rev."},{"key":"mlstae0d41bib23","doi-asserted-by":"publisher","first-page":"481","DOI":"10.1016\/S0009-2614(96)01449-2","type":"journal-article","article-title":"Atomic reference energies for density functional calculations","volume":"265","author":"Baerends","year":"1997","journal-title":"Chem. Phys. Lett."},{"key":"mlstae0d41bib24","doi-asserted-by":"publisher","first-page":"30","DOI":"10.1093\/nsr\/nwx105","type":"journal-article","article-title":"An overview of multi-task learning","volume":"5","author":"Zhang","year":"2017","journal-title":"Natl Sci. Rev."},{"key":"mlstae0d41bib25","doi-asserted-by":"publisher","first-page":"1","DOI":"10.1088\/2632-2153\/ad2cef","type":"journal-article","article-title":"CHGNet as a pretrained universal neural network potential for charge-informed atomistic modelling","author":"Deng","year":"2023","journal-title":"Nat. Mach. Intell."},{"key":"mlstae0d41bib26","doi-asserted-by":"publisher","first-page":"1066","DOI":"10.1021\/acs.jcim.0c01224","type":"journal-article","article-title":"Transferable multilevel attention neural network for accurate prediction of quantum chemistry properties via multitask learning","volume":"61","author":"Liu","year":"2021","journal-title":"J. Chem. Information Mod."},{"key":"mlstae0d41bib27","article-title":"Parameterized neural networks for finance","author":"Oeltz","year":"2023","type":"preprint"},{"key":"mlstae0d41bib28","doi-asserted-by":"publisher","DOI":"10.1002\/ange.20220573510.1002\/ange.202205735","type":"journal-article","article-title":"Best-practice dft protocols for basic molecular computational chemistry","volume":"134","author":"Bursch","year":"2022","journal-title":"Angew. Chem. Int. Ed. Engl."},{"key":"mlstae0d41bib29","doi-asserted-by":"publisher","first-page":"351","DOI":"10.3934\/acse.2023015","type":"journal-article","article-title":"Review of multi-fidelity models","volume":"1","author":"Fern\u00e1ndez-Godino","year":"2023","journal-title":"Adv. Comput. Sci. Eng."},{"key":"mlstae0d41bib30","doi-asserted-by":"crossref","DOI":"10.26434\/chemrxiv-2023-8n737","type":"other","article-title":"Leveraging multitask learning to improve the transferability of machine learned force fields","author":"Jacobson","year":"2023"},{"key":"mlstae0d41bib31","doi-asserted-by":"publisher","DOI":"10.1063\/1.5090222","type":"journal-article","article-title":"A generally applicable atomic-charge dependent London dispersion correction","volume":"150","author":"Caldeweyher","year":"2019","journal-title":"J. Chem. Phys."},{"key":"mlstae0d41bib32","doi-asserted-by":"publisher","first-page":"2326","DOI":"10.1021\/jp5065819","type":"journal-article","article-title":"Additive covalent radii for single-, double- and triple-bonded molecules and tetrahedrally bonded crystals: a summary","volume":"119","author":"Pyykk\u00f6","year":"2015","journal-title":"J. Phys. Chem. A"},{"key":"mlstae0d41bib33","doi-asserted-by":"publisher","DOI":"10.1063\/1.1904565","type":"journal-article","article-title":"Prescription for the design and selection of density functional approximations: more constraint satisfaction with fewer fits","volume":"123","author":"Perdew","year":"2005","journal-title":"J. Chem. Phys."},{"key":"mlstae0d41bib34","article-title":"A foundational potential energy surface dataset for materials","author":"Kaplan","year":"2025","type":"preprint"},{"key":"mlstae0d41bib35","doi-asserted-by":"publisher","first-page":"718","DOI":"10.1038\/s43588-022-00349-3","type":"journal-article","article-title":"A universal graph deep learning interatomic potential for the periodic table","volume":"2","author":"Chen","year":"2022","journal-title":"Nat. Comput. Sci."},{"key":"mlstae0d41bib36","author":"Rasmussen","year":"2006","type":"book"},{"key":"mlstae0d41bib37","article-title":"Multixc-qm9 data","author":"Nandi","year":"2023","type":"other"}],"container-title":["Machine Learning: Science and Technology"],"original-title":[],"link":[{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae0d41","content-type":"text\/html","content-version":"am","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae0d41\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae0d41","content-type":"text\/html","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae0d41\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae0d41\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"syndication"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae0d41\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"syndication"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae0d41\/pdf","content-type":"application\/pdf","content-version":"am","intended-application":"similarity-checking"},{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae0d41\/pdf","content-type":"application\/pdf","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,10,10]],"date-time":"2025-10-10T10:58:04Z","timestamp":1760093884000},"score":1,"resource":{"primary":{"URL":"https:\/\/iopscience.iop.org\/article\/10.1088\/2632-2153\/ae0d41"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,10,10]]},"references-count":37,"journal-issue":{"issue":"4","published-online":{"date-parts":[[2025,10,10]]},"published-print":{"date-parts":[[2025,12,30]]}},"URL":"https:\/\/doi.org\/10.1088\/2632-2153\/ae0d41","relation":{"has-preprint":[{"id-type":"doi","id":"10.26434\/chemrxiv-2025-vx7nx-v2","asserted-by":"object"}]},"ISSN":["2632-2153"],"issn-type":[{"value":"2632-2153","type":"electronic"}],"subject":[],"published":{"date-parts":[[2025,10,10]]},"assertion":[{"value":"Multi-fidelity learning for atomistic models via trainable data embeddings","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 2025 The Author(s). Published by IOP Publishing Ltd","name":"copyright_information","label":"Copyright Information"},{"value":"2025-05-19","name":"date_received","label":"Date Received","group":{"name":"publication_dates","label":"Publication dates"}},{"value":"2025-09-29","name":"date_accepted","label":"Date Accepted","group":{"name":"publication_dates","label":"Publication dates"}},{"value":"2025-10-10","name":"date_epub","label":"Online publication date","group":{"name":"publication_dates","label":"Publication dates"}}]}}