{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/less-is-more-sampling-chemical-space-with","title":"Less is more: sampling chemical space with active learning","arxiv_id":"1801.09319","date":"2018-01-28","proceeding":null,"authors":["Justin S. Smith","Ben Nebgen","Nicholas Lubbers","Olexandr Isayev","Adrian E. Roitberg"],"abstract":"The development of accurate and transferable machine learning (ML) potentials\nfor predicting molecular energetics is a challenging task. The process of data\ngeneration to train such ML potentials is a task neither well understood nor\nresearched in detail. In this work, we present a fully automated approach for\nthe generation of datasets with the intent of training universal ML potentials.\nIt is based on the concept of active learning (AL) via Query by Committee\n(QBC), which uses the disagreement between an ensemble of ML potentials to\ninfer the reliability of the ensemble's prediction. QBC allows the presented AL\nalgorithm to automatically sample regions of chemical space where the ML\npotential fails to accurately predict the potential energy. AL improves the\noverall fitness of ANAKIN-ME (ANI) deep learning potentials in rigorous test\ncases by mitigating human biases in deciding what new training data to use. AL\nalso reduces the training set size to a fraction of the data required when\nusing naive random sampling techniques. To provide validation of our AL\napproach we develop the COMP6 benchmark (publicly available on GitHub), which\ncontains a diverse set of organic molecules. Through the AL process, it is\nshown that the AL-based potentials perform as well as the ANI-1 potential on\nCOMP6 with only 10% of the data, and vastly outperforms ANI-1 with 25% the\namount of data. Finally, we show that our proposed AL technique develops a\nuniversal ANI potential (ANI-1x) that provides accurate energy and force\npredictions on the entire COMP6 benchmark. This universal ML potential achieves\na level of accuracy on par with the best ML potentials for single molecule or\nmaterials, while remaining applicable to the general class of organic molecules\ncomprised of the elements CHNO.","url_abs":"http://arxiv.org/abs/1801.09319v2","url_pdf":"http://arxiv.org/pdf/1801.09319v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"less-is-more-sampling-chemical-space-with","repo_url":"https://github.com/isayev/ASE_ANI","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"unanswered"}},{"paper_slug":"less-is-more-sampling-chemical-space-with","repo_url":"https://github.com/isayev/COMP6","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}},{"paper_slug":"less-is-more-sampling-chemical-space-with","repo_url":"https://github.com/isayev/ANI1_dataset","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":"active-learning","task_name":"Active Learning"}],"methods":[],"datasets_introduced":[{"slug":"comp6","name":"COMP6","full_name":"COmprehensive Machine-learning Potential"}],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/1801.09319","atlas_url":"https://app.syntology.ai/?focus=1801.09319","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}