{"id":"https://openalex.org/W4404782809","doi":"https://doi.org/10.18653/v1/2024.emnlp-main.965","title":"Information Flow Routes: Automatically Interpreting Language Models at Scale","display_name":"Information Flow Routes: Automatically Interpreting Language Models at Scale","publication_year":2024,"publication_date":"2024-01-01","ids":{"openalex":"https://openalex.org/W4404782809","doi":"https://doi.org/10.18653/v1/2024.emnlp-main.965"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2024.emnlp-main.965","doi":"https://doi.org/10.18653/v1/2024.emnlp-main.965","is_oa":true,"landing_page_url":"http://dx.doi.org/10.18653/v1/2024.emnlp-main.965","pdf_url":"https://aclanthology.org/2024.emnlp-main.965.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://aclanthology.org/2024.emnlp-main.965.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5110335380","display_name":"Javier Ferrando","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I9617848","display_name":"Universitat Polit\u00e8cnica de Catalunya","ror":"https://ror.org/03mb6wj31","country_code":"ES","type":"education","lineage":["https://openalex.org/I9617848"]}],"countries":["ES"],"is_corresponding":false,"raw_author_name":"Javier Ferrando","raw_affiliation_strings":["Universitat Polit\u00e8cnica de Catalunya"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Universitat Polit\u00e8cnica de Catalunya","institution_ids":["https://openalex.org/I9617848"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5077727226","display_name":"Elena Voita","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210114444","display_name":"Meta (United States)","ror":"https://ror.org/01zbnvs85","country_code":"US","type":"company","lineage":["https://openalex.org/I4210114444"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Elena Voita","raw_affiliation_strings":["Meta AI"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Meta AI","institution_ids":["https://openalex.org/I4210114444"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":1.7323,"has_fulltext":true,"cited_by_count":6,"citation_normalized_percentile":{"value":0.8722187,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":94,"max":99},"biblio":{"volume":null,"issue":null,"first_page":"17432","last_page":"17445"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.9970999956130981,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.9970999956130981,"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":0.001500000013038516,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.0003000000142492354,"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-model-interpretability","display_name":"language model interpretability","score":0.6690000295639038},{"id":"https://openalex.org/keywords/attribution","display_name":"attribution","score":0.47699999809265137},{"id":"https://openalex.org/keywords/llama-2","display_name":"Llama 2","score":0.8289999961853027},{"id":"https://openalex.org/keywords/domain-specialization","display_name":"domain specialization","score":0.7080000042915344},{"id":"https://openalex.org/keywords/model-behavior-analysis","display_name":"model behavior analysis","score":0.40299999713897705}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.799616277217865},{"id":"https://openalex.org/C2779136372","wikidata":"https://www.wikidata.org/wiki/Q10283002","display_name":"Information flow","level":2,"score":0.5753044486045837},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.5453037023544312},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.543351948261261},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.3757588267326355},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.37449151277542114},{"id":"https://openalex.org/C23123220","wikidata":"https://www.wikidata.org/wiki/Q816826","display_name":"Information retrieval","level":1,"score":0.35909920930862427},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.14508968591690063},{"id":"https://openalex.org/C58640448","wikidata":"https://www.wikidata.org/wiki/Q42515","display_name":"Cartography","level":1,"score":0.06604161858558655},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.06281983852386475},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.0}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.18653/v1/2024.emnlp-main.965","doi":"https://doi.org/10.18653/v1/2024.emnlp-main.965","is_oa":true,"landing_page_url":"http://dx.doi.org/10.18653/v1/2024.emnlp-main.965","pdf_url":"https://aclanthology.org/2024.emnlp-main.965.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2403.00824","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2403.00824","pdf_url":"https://arxiv.org/pdf/2403.00824","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2403.00824","doi":"https://doi.org/10.48550/arxiv.2403.00824","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2403.00824","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.18653/v1/2024.emnlp-main.965","doi":"https://doi.org/10.18653/v1/2024.emnlp-main.965","is_oa":true,"landing_page_url":"http://dx.doi.org/10.18653/v1/2024.emnlp-main.965","pdf_url":"https://aclanthology.org/2024.emnlp-main.965.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4404782809.pdf","grobid_xml":"https://content.openalex.org/works/W4404782809.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7214098872","https://openalex.org/W7221811189","https://openalex.org/W7092296131","https://openalex.org/W7164194511","https://openalex.org/W7166832719","https://openalex.org/W3101068231","https://openalex.org/W7214616596","https://openalex.org/W7172506373","https://openalex.org/W3204495782","https://openalex.org/W7162314614"],"abstract_inverted_index":{"Information":[0],"flows":[1],"by":[2],"routes":[3,13,95],"inside":[4],"the":[5,11,42,50,101,104,156],"network":[6],"via":[7],"mechanisms":[8],"implemented":[9],"in":[10,33,114,148],"model.These":[12],"can":[14,91,110,167],"be":[15,168],"represented":[16],"as":[17,173],"graphs":[18,32],"where":[19],"nodes":[20,45],"correspond":[21],"to":[22,27,49,65,84],"token":[23,139],"representations":[24],"and":[25,46,89,128,141],"edges":[26],"computations.We":[28],"automatically":[29],"build":[30],"these":[31],"a":[34,72,82,107],"top-down":[35],"manner,":[36],"for":[37,96,116],"each":[38],"prediction":[39,87,98],"leaving":[40],"only":[41],"most":[43],"important":[44],"edges.In":[47],"contrast":[48],"existing":[51,68],"workflows":[52],"relying":[53],"on":[54,170],"activation":[55],"patching,":[56,77],"we":[57,78,90,109,145,161],"do":[58,79],"this":[59,62],"through":[60],"attribution:":[61],"allows":[63],"us":[64],"efficiently":[66],"uncover":[67],"circuits":[69],"with":[70,76,125],"just":[71,100],"single":[73],"forward":[74],"pass.Unlike":[75],"not":[80],"need":[81],"human":[83],"carefully":[85],"design":[86],"templates,":[88],"extract":[92],"information":[93],"flow":[94],"any":[97],"(not":[99],"ones":[102],"among":[103],"allowed":[105],"templates).As":[106],"result,":[108],"analyze":[111],"model":[112,165],"behavior":[113,151],"general,":[115],"specific":[117],"types":[118],"of":[119,155,159],"predictions,":[120],"or":[121,175],"different":[122],"domains.We":[123],"experiment":[124],"Llama":[126,149],"2":[127,150],"show":[129,162],"that":[130,163],"some":[131,164],"attention":[132],"head":[133],"roles":[134],"are":[135],"overall":[136],"important,":[137],"e.g.previous":[138],"heads":[140],"subword":[142],"merging":[143],"heads.Next,":[144],"find":[146],"similarities":[147],"when":[152],"handling":[153],"tokens":[154],"same":[157],"part":[158],"speech.Finally,":[160],"components":[166],"specialized":[169],"domains":[171],"such":[172],"coding":[174],"multilingual":[176],"texts.":[177]},"counts_by_year":[{"year":2026,"cited_by_count":4},{"year":2025,"cited_by_count":2}],"updated_date":"2026-10-14T16:39:01.820549","created_date":"2024-11-28T00:00:00"}