{"id":"https://openalex.org/W4400518732","doi":"https://doi.org/10.48550/arxiv.2407.06483","title":"Composable Interventions for Language Models","display_name":"Composable Interventions for Language Models","publication_year":2024,"publication_date":"2024-07-09","ids":{"openalex":"https://openalex.org/W4400518732","doi":"https://doi.org/10.48550/arxiv.2407.06483"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2407.06483","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2407.06483","pdf_url":"https://arxiv.org/pdf/2407.06483","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2407.06483","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5124872126","display_name":"Arinbj\u00f6rn Kolbeinsson","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I51556381","display_name":"University of Virginia","ror":"https://ror.org/0153tk833","country_code":"US","type":"education","lineage":["https://openalex.org/I51556381"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Kolbeinsson, Arinbjorn","raw_affiliation_strings":["University of Virginia & Askan"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Virginia & Askan","institution_ids":["https://openalex.org/I51556381"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5123246929","display_name":"Kyle O'Brien","orcid":"https://orcid.org/0009-0005-8219-1617","observed_orcids":["https://orcid.org/0009-0005-8219-1617"]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"O'Brien, Kyle","raw_affiliation_strings":["University of Oxford"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Oxford","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"middle","author":{"id":null,"display_name":"Huang, Tianjin","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Tianjin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111361941","display_name":"Shanghua Gao","orcid":"https://orcid.org/0000-0002-7055-2703","observed_orcids":["https://orcid.org/0000-0002-7055-2703"]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Gao, Shanghua","raw_affiliation_strings":["Harvard Medical School"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harvard Medical School","institution_ids":["https://openalex.org/I136199984"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5107846414","display_name":"Shiwei Liu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I40120149","display_name":"University of Oxford","ror":"https://ror.org/052gg0110","country_code":"GB","type":"education","lineage":["https://openalex.org/I40120149"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Liu, Shiwei","raw_affiliation_strings":["University of Oxford"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Oxford","institution_ids":["https://openalex.org/I40120149"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5012775658","display_name":"Jonathan Richard Schwarz","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Schwarz, Jonathan Richard","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5023061511","display_name":"Anurag Jayant Vaidya","orcid":"https://orcid.org/0000-0001-6378-6010","observed_orcids":["https://orcid.org/0000-0001-6378-6010"]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]},{"id":"https://openalex.org/I48633490","display_name":"Mass General Brigham","ror":"https://ror.org/04py2rh25","country_code":"US","type":"healthcare","lineage":["https://openalex.org/I48633490"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Vaidya, Anurag","raw_affiliation_strings":["Harvard Medical School Mass General Brigham"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harvard Medical School Mass General Brigham","institution_ids":["https://openalex.org/I136199984","https://openalex.org/I48633490"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5080050834","display_name":"Faisal Mahmood","orcid":"https://orcid.org/0000-0001-7587-1562","observed_orcids":["https://orcid.org/0000-0001-7587-1562"]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]},{"id":"https://openalex.org/I4210111299","display_name":"Thomson Reuters Foundation","ror":"https://ror.org/01xpmbc27","country_code":"US","type":"nonprofit","lineage":["https://openalex.org/I4210111299"]},{"id":"https://openalex.org/I4210132832","display_name":"Thomson Reuters Foundation","ror":"https://ror.org/040j2ng64","country_code":"GB","type":"other","lineage":["https://openalex.org/I4210132832"]},{"id":"https://openalex.org/I48633490","display_name":"Mass General Brigham","ror":"https://ror.org/04py2rh25","country_code":"US","type":"healthcare","lineage":["https://openalex.org/I48633490"]},{"id":"https://openalex.org/I51556381","display_name":"University of Virginia","ror":"https://ror.org/0153tk833","country_code":"US","type":"education","lineage":["https://openalex.org/I51556381"]}],"countries":["GB","US"],"is_corresponding":false,"raw_author_name":"Mahmood, Faisal","raw_affiliation_strings":["Harvard Medical School Mass General Brigham","University of Virginia & Thomson-Reuters Foundational Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harvard Medical School Mass General Brigham","institution_ids":["https://openalex.org/I136199984","https://openalex.org/I48633490"]},{"raw_affiliation_string":"University of Virginia & Thomson-Reuters Foundational Research","institution_ids":["https://openalex.org/I4210111299","https://openalex.org/I4210132832","https://openalex.org/I51556381"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5086052373","display_name":"Marinka \u017ditnik","orcid":"https://orcid.org/0000-0001-8530-7228","observed_orcids":["https://orcid.org/0000-0001-8530-7228"]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Zitnik, Marinka","raw_affiliation_strings":["Harvard Medical School"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harvard Medical School","institution_ids":["https://openalex.org/I136199984"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5103073431","display_name":"Tianlong Chen","orcid":"https://orcid.org/0000-0001-7774-8197","observed_orcids":["https://orcid.org/0000-0001-7774-8197","https://orcid.org/0000-0001-5588-717X"]},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Chen, Tianlong","raw_affiliation_strings":["UNC Chapel Hill"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UNC Chapel Hill","institution_ids":["https://openalex.org/I114027177"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5075881948","display_name":"Thomas Hartvigsen","orcid":"https://orcid.org/0000-0002-5288-2792","observed_orcids":["https://orcid.org/0000-0002-5288-2792"]},"institutions":[{"id":"https://openalex.org/I136199984","display_name":"Harvard University","ror":"https://ror.org/03vek6s52","country_code":"US","type":"education","lineage":["https://openalex.org/I136199984"]},{"id":"https://openalex.org/I48633490","display_name":"Mass General Brigham","ror":"https://ror.org/04py2rh25","country_code":"US","type":"healthcare","lineage":["https://openalex.org/I48633490"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hartvigsen, Thomas","raw_affiliation_strings":["Harvard Medical School Mass General Brigham"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Harvard Medical School Mass General Brigham","institution_ids":["https://openalex.org/I136199984","https://openalex.org/I48633490"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":7,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":6,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.3086000084877014,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.3086000084877014,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.2797999978065491,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.1859000027179718,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/language-models","display_name":"language models","score":0.7490000128746033},{"id":"https://openalex.org/keywords/knowledge-editing","display_name":"knowledge editing","score":0.9330000281333923},{"id":"https://openalex.org/keywords/model-compression","display_name":"model compression","score":0.9819999933242798},{"id":"https://openalex.org/keywords/machine-unlearning","display_name":"machine unlearning","score":0.9990000128746033},{"id":"https://openalex.org/keywords/intervention-orders","display_name":"intervention orders","score":0.7400000095367432},{"id":"https://openalex.org/keywords/factual-accuracy","display_name":"factual accuracy","score":0.7379999756813049}],"concepts":[{"id":"https://openalex.org/C27415008","wikidata":"https://www.wikidata.org/wiki/Q7256382","display_name":"Psychological intervention","level":2,"score":0.5243209600448608},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5156549215316772},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.4042694866657257},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.3694803714752197},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.24693620204925537},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.13298344612121582},{"id":"https://openalex.org/C118552586","wikidata":"https://www.wikidata.org/wiki/Q7867","display_name":"Psychiatry","level":1,"score":0.0}],"mesh":[],"locations_count":4,"locations":[{"id":"pmh:oai:arXiv.org:2407.06483","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2407.06483","pdf_url":"https://arxiv.org/pdf/2407.06483","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"pmh:oai:figshare.com:article/29843924","doi":null,"is_oa":false,"landing_page_url":"https://figshare.com/articles/conference_contribution/Composable_Interventions_for_Language_Models/29843924","pdf_url":null,"source":{"id":"https://openalex.org/S4377196282","display_name":"Figshare","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I4210132348","host_organization_name":"Figshare (United Kingdom)","host_organization_lineage":["https://openalex.org/I4210132348"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Conference contribution"},{"id":"pmh:oai:ore.exeter.ac.uk:10871/140593","doi":null,"is_oa":false,"landing_page_url":"http://hdl.handle.net/10871/140593","pdf_url":null,"source":{"id":"https://openalex.org/S4306401998","display_name":"Open Research Exeter (University of Exeter)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I23923803","host_organization_name":"University of Exeter","host_organization_lineage":["https://openalex.org/I23923803"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Conference paper"},{"id":"doi:10.48550/arxiv.2407.06483","doi":"https://doi.org/10.48550/arxiv.2407.06483","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2407.06483","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2407.06483","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2407.06483","pdf_url":"https://arxiv.org/pdf/2407.06483","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4400518732.pdf","grobid_xml":"https://content.openalex.org/works/W4400518732.grobid-xml"},"referenced_works_count":32,"referenced_works":["https://openalex.org/W4320005767","https://openalex.org/W4322825541","https://openalex.org/W4361866125","https://openalex.org/W4368755730","https://openalex.org/W4379260375","https://openalex.org/W4380994473","https://openalex.org/W4381586827","https://openalex.org/W4383174296","https://openalex.org/W4384920109","https://openalex.org/W4385570473","https://openalex.org/W4385570867","https://openalex.org/W4385571289","https://openalex.org/W4385572016","https://openalex.org/W4385849354","https://openalex.org/W4387356300","https://openalex.org/W4387559767","https://openalex.org/W4390136566","https://openalex.org/W4392223220","https://openalex.org/W4396987412","https://openalex.org/W4402669768","https://openalex.org/W4402670091","https://openalex.org/W4402704595","https://openalex.org/W4403195962","https://openalex.org/W4404708545","https://openalex.org/W4404781851","https://openalex.org/W4404782881","https://openalex.org/W4404918643","https://openalex.org/W4406152279","https://openalex.org/W4415800792","https://openalex.org/W6941597699","https://openalex.org/W7133198810","https://openalex.org/W7133245307"],"related_works":["https://openalex.org/W7160505814","https://openalex.org/W7212072200","https://openalex.org/W7204932479","https://openalex.org/W4417119681","https://openalex.org/W4407386357","https://openalex.org/W4404404507","https://openalex.org/W7167813011","https://openalex.org/W7128000853","https://openalex.org/W4415798367","https://openalex.org/W4407245244"],"abstract_inverted_index":{"Test-time":[0],"interventions":[1,29,37,54,73,125],"for":[2,138,153],"language":[3,77],"models":[4],"can":[5],"enhance":[6],"factual":[7],"accuracy,":[8],"mitigate":[9],"harmful":[10],"outputs,":[11],"and":[12,82,93,107,122,132],"improve":[13],"model":[14],"efficiency":[15],"without":[16],"costly":[17],"retraining.":[18],"But":[19],"despite":[20],"a":[21,64,83,151],"flood":[22],"of":[23,28,70,130,158],"new":[24,80,154],"methods,":[25],"different":[26,114],"types":[27],"are":[30,136],"largely":[31],"developing":[32],"independently.":[33],"In":[34],"practice,":[35],"multiple":[36,72],"must":[38],"be":[39],"applied":[40],"sequentially":[41],"to":[42,51,66],"the":[43,68,75],"same":[44,76],"model,":[45],"yet":[46],"we":[47,89],"lack":[48],"standardized":[49],"ways":[50],"study":[52,67],"how":[53],"interact.":[55],"We":[56],"fill":[57],"this":[58],"gap":[59],"by":[60],"introducing":[61],"composable":[62],"interventions,":[63],"framework":[65],"effects":[69],"using":[71],"on":[74,127],"models,":[78],"featuring":[79],"metrics":[81,135],"unified":[84],"codebase.":[85],"Using":[86],"our":[87,143,159],"framework,":[88],"conduct":[90],"extensive":[91],"experiments":[92],"compose":[94],"popular":[95,133],"methods":[96],"from":[97,112],"three":[98],"emerging":[99],"intervention":[100],"categories":[101],"--":[102],"Knowledge":[103],"Editing,":[104],"Model":[105],"Compression,":[106],"Machine":[108],"Unlearning.":[109],"Our":[110],"results":[111],"310":[113],"compositions":[115],"uncover":[116],"meaningful":[117],"interactions:":[118],"compression":[119],"hinders":[120],"editing":[121],"unlearning,":[123],"composing":[124],"hinges":[126],"their":[128],"order":[129],"application,":[131],"general-purpose":[134],"inadequate":[137],"assessing":[139],"composability.":[140],"Taken":[141],"together,":[142],"findings":[144],"showcase":[145],"clear":[146],"gaps":[147],"in":[148],"composability,":[149],"suggesting":[150],"need":[152],"multi-objective":[155],"interventions.":[156],"All":[157],"code":[160],"is":[161],"public:":[162],"https://github.com/hartvigsen-group/composable-interventions.":[163]},"counts_by_year":[{"year":2025,"cited_by_count":3},{"year":2024,"cited_by_count":3}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2024-07-11T00:00:00"}
