{"corpus_id":195820364,"paper_sha":"753b7a701adc1b6072378bd048cfa8567885d9c7","doi":null,"arxiv_id":"1907.02893","pmid":null,"pmcid":null,"mag_id":2953494151,"dblp_id":"journals/corr/abs-1907-02893","acl_id":null,"title":"Invariant Risk Minimization","year":2019,"publication_date":"2019-07-05","venue":"arXiv.org","journal":{"name":"ArXiv","pages":null,"volume":"abs/1907.02893"},"journal_issn":null,"journal_title":null,"publication_types":["JournalArticle"],"pubmed_pub_types":null,"s2_fields_of_study":["Mathematics","Computer Science"],"reference_count":77,"citation_count":2635,"influential_citation_count":594,"is_open_access":false,"arxiv_categories":["stat.ML","cs.AI","cs.LG"],"arxiv_license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","arxiv_journal_ref":null,"mesh_headings":null,"chemicals":null,"comments_corrections":null,"source_flags":1,"s2_open_access_pdf_url":null,"s2_open_access_landing_url":null,"s2_open_access_license":null,"s2_open_access_status":null,"pmc_open_access_pdf_url":null,"pmc_open_access_landing_url":null,"pmc_open_access_license":null,"pmc_open_access_status":null,"unpaywall_open_access_pdf_url":null,"unpaywall_open_access_landing_url":null,"unpaywall_open_access_license":null,"unpaywall_open_access_status":null,"abstract":"We introduce Invariant Risk Minimization (IRM), a learning paradigm to estimate invariant correlations across multiple training distributions. To achieve this goal, IRM learns a data representation such that the optimal classifier, on top of that data representation, matches for all training distributions. Through theory and experiments, we show how the invariances learned by IRM relate to the causal structures governing the data and enable out-of-distribution generalization.","claims":[{"public_id":"cl_9dbd24c98d16b57733d38c8100b3ab51","status":"active","text":"Invariant Risk Minimization enables out-of-distribution generalization.","confidence":0.96,"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/claims/cl_9dbd24c98d16b57733d38c8100b3ab51"},{"public_id":"cl_f1907e8272160f4ff684bb799b9f9e9c","status":"active","text":"Invariant Risk Minimization estimates invariant correlations across multiple training distributions by learning a representation whose optimal classifier is the same for all training distributions.","confidence":0.98,"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/claims/cl_f1907e8272160f4ff684bb799b9f9e9c"},{"public_id":"cl_92c4124bc23a8a6b40bc1d638423c660","status":"active","text":"The invariances learned by Invariant Risk Minimization relate to the causal structures governing the data.","confidence":0.95,"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/claims/cl_92c4124bc23a8a6b40bc1d638423c660"}],"concepts":[{"public_id":"co_10b53d031bfe02a98408d7b21b4d9e99","status":"active","name":"Invariant Risk Minimization","description":"A learning paradigm that seeks representations whose predictive relationship with the target is invariant across training environments.","types":["learning paradigm","method"],"aliases":["IRM"],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_10b53d031bfe02a98408d7b21b4d9e99"},{"public_id":"co_1d31af25fc7d9029f0dceaeb43244679","status":"active","name":"optimal classifier","description":"The classifier that minimizes risk on top of a given representation within each training distribution.","types":["model"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_1d31af25fc7d9029f0dceaeb43244679"},{"public_id":"co_6515bf308a6b24767f73979dc8956c93","status":"active","name":"training distributions","description":"The distributions on which the model is trained and across which invariance is enforced.","types":["distribution"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_6515bf308a6b24767f73979dc8956c93"},{"public_id":"co_682610ffd7239dd570a05d0c07a3dc30","status":"active","name":"causal structures","description":"Underlying causal relationships that govern how variables in the data are generated.","types":["causal model"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_682610ffd7239dd570a05d0c07a3dc30"},{"public_id":"co_76bb78c6e7fd86da1424c66ef31d1361","status":"active","name":"multiple training distributions","description":"More than one training environment or data distribution used to learn a model.","types":["training setting"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_76bb78c6e7fd86da1424c66ef31d1361"},{"public_id":"co_ad04a6d1a091ecf05ed116c319768ca1","status":"active","name":"invariances","description":"Properties of the learned predictor that remain stable across environments.","types":["property"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_ad04a6d1a091ecf05ed116c319768ca1"},{"public_id":"co_c128cf15d1264f60df4a2479ad631514","status":"active","name":"invariant correlations","description":"Predictive relationships that remain stable across different training distributions.","types":["statistical property"],"aliases":[],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_c128cf15d1264f60df4a2479ad631514"},{"public_id":"co_d0829620cde17492d52d96f9f9abeeb0","status":"active","name":"out-of-distribution generalization","description":"The ability of a learned model to perform well on data drawn from distributions different from the training data.","types":["generalization capability"],"aliases":["OOD generalization"],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_d0829620cde17492d52d96f9f9abeeb0"},{"public_id":"co_d2d766b91caa87d17c1aabdba4fae0c7","status":"active","name":"data representation","description":"A learned feature mapping used as input to a classifier.","types":["model component"],"aliases":["representation"],"contributors":[{"id":1,"public_id":"12632b8b5f","public_label":"Anonymous (12632b8b5f)","roles":["extraction"],"url":"https://sah.borca.ai/u/12632b8b5f"}],"url":"https://sah.borca.ai/concepts/co_d2d766b91caa87d17c1aabdba4fae0c7"}],"external_ids":{"DOI":null,"ArXiv":"1907.02893","PubMed":null,"PubMedCentral":null,"MAG":2953494151,"DBLP":"journals/corr/abs-1907-02893","ACL":null},"open_access":{"is_open_access":true,"pdf_url":"https://arxiv.org/pdf/1907.02893","landing_url":"https://arxiv.org/abs/1907.02893","source":"arxiv","pdf_url_source":"derived_arxiv","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","reason":null},"reference_availability":{"status":"available","references_indexed":true,"full_text_available":true,"full_text_source":"arxiv","count_basis":"semantic_scholar_metadata","extraction_status":"not_applicable","reason":null},"source":{"provider":"episteme2","base_corpus":"semantic_scholar_dump","freshness_mode":"unknown","basis":["semantic_scholar_metadata","postgres_metadata"],"limits":["paper metadata is based on indexed upstream scholarly datasets","claims and concepts are available only for extracted papers","absence of claims or concepts means no extracted graph data is available in this response"],"status":"available","degraded":false,"degraded_reasons":[],"diagnostics":{"status":"available","degraded":false,"degraded_reasons":[],"metadata_status":"available","graph_status":"available","abstract_status":"available"},"source_flags":1},"paper_id":634745,"paper_uid":"6911f5e0-e9a8-46c6-ac7e-2d94ef30d8fd","canonical_identity":{"paper_id":634745,"paper_uid":"6911f5e0-e9a8-46c6-ac7e-2d94ef30d8fd","identity_status":"available","lookup_basis":"semantic_scholar_external_id","compatibility_path":"corpus_id"},"url":"https://sah.borca.ai/papers/195820364"}