{"id":"https://openalex.org/W4389518693","doi":"https://doi.org/10.18653/v1/2023.emnlp-main.403","title":"Understanding the Inner-workings of Language Models Through Representation Dissimilarity","display_name":"Understanding the Inner-workings of Language Models Through Representation Dissimilarity","publication_year":2023,"publication_date":"2023-01-01","ids":{"openalex":"https://openalex.org/W4389518693","doi":"https://doi.org/10.18653/v1/2023.emnlp-main.403"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2023.emnlp-main.403","doi":"https://doi.org/10.18653/v1/2023.emnlp-main.403","is_oa":true,"landing_page_url":"http://dx.doi.org/10.18653/v1/2023.emnlp-main.403","pdf_url":"https://aclanthology.org/2023.emnlp-main.403.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 2023 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","is_traditional":true,"indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://aclanthology.org/2023.emnlp-main.403.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5008756360","display_name":"Davis Brown","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I142606810","display_name":"Pacific Northwest National Laboratory","ror":"https://ror.org/05h992307","country_code":"US","type":"facility","lineage":["https://openalex.org/I1325736334","https://openalex.org/I1330989302","https://openalex.org/I142606810","https://openalex.org/I39565521"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Davis Brown","raw_affiliation_strings":["Pacific Northwest National Laboratory"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Pacific Northwest National Laboratory","institution_ids":["https://openalex.org/I142606810"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5024921721","display_name":"Charles Godfrey","orcid":"https://orcid.org/0000-0003-1698-2718","observed_orcids":["https://orcid.org/0000-0003-1698-2718"]},"institutions":[{"id":"https://openalex.org/I68384125","display_name":"Thomson Reuters (United States)","ror":"https://ror.org/00m7gt169","country_code":"US","type":"company","lineage":["https://openalex.org/I68384125"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Charles Godfrey","raw_affiliation_strings":["Thomson Reuters Labs"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Thomson Reuters Labs","institution_ids":["https://openalex.org/I68384125"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5067091221","display_name":"Nicholas Konz","orcid":"https://orcid.org/0000-0003-0230-1598","observed_orcids":["https://orcid.org/0000-0003-0230-1598"]},"institutions":[{"id":"https://openalex.org/I142606810","display_name":"Pacific Northwest National Laboratory","ror":"https://ror.org/05h992307","country_code":"US","type":"facility","lineage":["https://openalex.org/I1325736334","https://openalex.org/I1330989302","https://openalex.org/I142606810","https://openalex.org/I39565521"]},{"id":"https://openalex.org/I170897317","display_name":"Duke University","ror":"https://ror.org/00py81415","country_code":"US","type":"education","lineage":["https://openalex.org/I170897317"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Nicholas Konz","raw_affiliation_strings":["Duke University","Pacific Northwest National Laboratory"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Duke University","institution_ids":["https://openalex.org/I170897317"]},{"raw_affiliation_string":"Pacific Northwest National Laboratory","institution_ids":["https://openalex.org/I142606810"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5065254827","display_name":"Jonathan H. Tu","orcid":"https://orcid.org/0000-0001-5450-6451","observed_orcids":["https://orcid.org/0000-0001-5450-6451"]},"institutions":[{"id":"https://openalex.org/I142606810","display_name":"Pacific Northwest National Laboratory","ror":"https://ror.org/05h992307","country_code":"US","type":"facility","lineage":["https://openalex.org/I1325736334","https://openalex.org/I1330989302","https://openalex.org/I142606810","https://openalex.org/I39565521"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Jonathan Tu","raw_affiliation_strings":["Pacific Northwest National Laboratory"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Pacific Northwest National Laboratory","institution_ids":["https://openalex.org/I142606810"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5066361433","display_name":"Henry Kvinge","orcid":"https://orcid.org/0000-0003-4108-1364","observed_orcids":["https://orcid.org/0000-0003-4108-1364"]},"institutions":[{"id":"https://openalex.org/I142606810","display_name":"Pacific Northwest National Laboratory","ror":"https://ror.org/05h992307","country_code":"US","type":"facility","lineage":["https://openalex.org/I1325736334","https://openalex.org/I1330989302","https://openalex.org/I142606810","https://openalex.org/I39565521"]},{"id":"https://openalex.org/I201448701","display_name":"University of Washington","ror":"https://ror.org/00cvxb145","country_code":"US","type":"education","lineage":["https://openalex.org/I201448701"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Henry Kvinge","raw_affiliation_strings":["Pacific Northwest National Laboratory","University of Washington"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Pacific Northwest National Laboratory","institution_ids":["https://openalex.org/I142606810"]},{"raw_affiliation_string":"University of Washington","institution_ids":["https://openalex.org/I201448701"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.7934,"has_fulltext":true,"cited_by_count":7,"citation_normalized_percentile":{"value":0.77976554,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":89,"max":97},"biblio":{"volume":null,"issue":null,"first_page":"6543","last_page":"6558"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.9146000146865845,"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":0.9146000146865845,"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":0.06840000301599503,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.002899999963119626,"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":0.9150000214576721},{"id":"https://openalex.org/keywords/internal-representations","display_name":"internal representations","score":0.7020000219345093},{"id":"https://openalex.org/keywords/interpretability","display_name":"interpretability","score":0.8610000014305115},{"id":"https://openalex.org/keywords/model-trust","display_name":"model trust","score":0.6970000267028809},{"id":"https://openalex.org/keywords/gelu-activation","display_name":"GELU activation","score":0.9980000257492065},{"id":"https://openalex.org/keywords/generalization","display_name":"generalization","score":0.8149999976158142},{"id":"https://openalex.org/keywords/model-interpretability","display_name":"model interpretability","score":0.4440000057220459}],"concepts":[{"id":"https://openalex.org/C2781067378","wikidata":"https://www.wikidata.org/wiki/Q17027399","display_name":"Interpretability","level":2,"score":0.9217091798782349},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6981779336929321},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.6673401594161987},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.663284182548523},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.6567416787147522},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.6388242840766907},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.5842428803443909},{"id":"https://openalex.org/C2780233690","wikidata":"https://www.wikidata.org/wiki/Q535347","display_name":"Transparency (behavior)","level":2,"score":0.532574474811554},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.5083536505699158},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4884625971317291},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.32170552015304565},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.20883554220199585},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.1800927221775055},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.07230743765830994},{"id":"https://openalex.org/C199539241","wikidata":"https://www.wikidata.org/wiki/Q7748","display_name":"Law","level":1,"score":0.0},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.0},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.0},{"id":"https://openalex.org/C17744445","wikidata":"https://www.wikidata.org/wiki/Q36442","display_name":"Political science","level":0,"score":0.0},{"id":"https://openalex.org/C94625758","wikidata":"https://www.wikidata.org/wiki/Q7163","display_name":"Politics","level":2,"score":0.0}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.18653/v1/2023.emnlp-main.403","doi":"https://doi.org/10.18653/v1/2023.emnlp-main.403","is_oa":true,"landing_page_url":"http://dx.doi.org/10.18653/v1/2023.emnlp-main.403","pdf_url":"https://aclanthology.org/2023.emnlp-main.403.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 2023 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2310.14993","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2310.14993","pdf_url":"https://arxiv.org/pdf/2310.14993","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":"doi:10.48550/arxiv.2310.14993","doi":"https://doi.org/10.48550/arxiv.2310.14993","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2310.14993","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":"doi:10.18653/v1/2023.emnlp-main.403","doi":"https://doi.org/10.18653/v1/2023.emnlp-main.403","is_oa":true,"landing_page_url":"http://dx.doi.org/10.18653/v1/2023.emnlp-main.403","pdf_url":"https://aclanthology.org/2023.emnlp-main.403.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 2023 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G125460123","display_name":null,"funder_award_id":"DE-AC05\u201376RL01830","funder_id":"https://openalex.org/F4320306084","funder_display_name":"U.S. Department of Energy"},{"id":"https://openalex.org/G138090931","display_name":null,"funder_award_id":"DEAC0576RL01830","funder_id":"https://openalex.org/F4320338354","funder_display_name":"Pacific Northwest National Laboratory"},{"id":"https://openalex.org/G2270264153","display_name":null,"funder_award_id":"DEAC05\u201376RL01830","funder_id":"https://openalex.org/F4320337547","funder_display_name":"Laboratory Directed Research and Development"},{"id":"https://openalex.org/G2988778103","display_name":null,"funder_award_id":"DE-AC05?","funder_id":"https://openalex.org/F4320306084","funder_display_name":"U.S. Department of Energy"},{"id":"https://openalex.org/G391136020","display_name":null,"funder_award_id":"DE- AC05-76RL01830","funder_id":"https://openalex.org/F4320306250","funder_display_name":"Battelle"},{"id":"https://openalex.org/G5269527469","display_name":null,"funder_award_id":"DE-AC05-76RL0","funder_id":"https://openalex.org/F4320306084","funder_display_name":"U.S. Department of Energy"},{"id":"https://openalex.org/G5726405315","display_name":null,"funder_award_id":"DE-AC05","funder_id":"https://openalex.org/F4320306250","funder_display_name":"Battelle"}],"funders":[{"id":"https://openalex.org/F4320306084","display_name":"U.S. Department of Energy","ror":"https://ror.org/01bj3aw27"},{"id":"https://openalex.org/F4320306250","display_name":"Battelle","ror":"https://ror.org/01h5tnr73"},{"id":"https://openalex.org/F4320337547","display_name":"Laboratory Directed Research and Development","ror":"https://ror.org/01e41cf67"},{"id":"https://openalex.org/F4320338354","display_name":"Pacific Northwest National Laboratory","ror":"https://ror.org/05h992307"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4389518693.pdf","grobid_xml":"https://content.openalex.org/works/W4389518693.grobid-xml"},"referenced_works_count":32,"referenced_works":["https://openalex.org/W2515741950","https://openalex.org/W2605717780","https://openalex.org/W2607892599","https://openalex.org/W2922565841","https://openalex.org/W2942810103","https://openalex.org/W2945976633","https://openalex.org/W2949074271","https://openalex.org/W2951286828","https://openalex.org/W2970820321","https://openalex.org/W2970862333","https://openalex.org/W2979826702","https://openalex.org/W2996074092","https://openalex.org/W3095872829","https://openalex.org/W3099624838","https://openalex.org/W3102226577","https://openalex.org/W3118781290","https://openalex.org/W3166956420","https://openalex.org/W3195830874","https://openalex.org/W3200668485","https://openalex.org/W3209091531","https://openalex.org/W3214698142","https://openalex.org/W4221155125","https://openalex.org/W4226128397","https://openalex.org/W4281681074","https://openalex.org/W4295689550","https://openalex.org/W4307103700","https://openalex.org/W4309217994","https://openalex.org/W4327526719","https://openalex.org/W4362655426","https://openalex.org/W4378474162","https://openalex.org/W4409242850","https://openalex.org/W7133214732"],"related_works":["https://openalex.org/W4399837584","https://openalex.org/W7221762642","https://openalex.org/W7223030650","https://openalex.org/W7162322464","https://openalex.org/W4389469960","https://openalex.org/W7221789202","https://openalex.org/W4402671619","https://openalex.org/W4416035258","https://openalex.org/W7171739978","https://openalex.org/W7202347467"],"abstract_inverted_index":{"As":[0],"language":[1,61,109,138],"models":[2,94],"are":[3,36,96,117,124],"applied":[4],"to":[5,42],"an":[6,18,67],"increasing":[7],"number":[8],"of":[9,60,74,93,107,128,137],"real-world":[10],"applications,":[11],"understanding":[12],"their":[13],"inner":[14,135],"workings":[15,136],"has":[16],"become":[17],"important":[19],"issue":[20],"in":[21,70],"model":[22,75,110],"trust,":[23],"interpretability,":[24],"and":[25,78,89,103,115],"transparency.In":[26],"this":[27],"work":[28],"we":[29],"show":[30],"that":[31,38,84,95,121],"representation":[32],"dissimilarity":[33,85,122],"measures,":[34],"which":[35,43],"functions":[37],"measure":[39],"the":[40,58,71,134],"extent":[41],"two":[44],"model's":[45],"internal":[46,72],"representations":[47,73],"differ,":[48],"can":[49,87],"be":[50],"a":[51,125],"valuable":[52],"tool":[53],"for":[54,130],"gaining":[55],"insight":[56],"into":[57],"mechanics":[59],"models.Among":[62],"our":[63],"insights":[64],"are:":[65],"(i)":[66],"apparent":[68],"asymmetry":[69],"using":[76],"SoLU":[77],"GeLU":[79],"activation":[80],"functions,":[81],"(ii)":[82],"evidence":[83],"measures":[86,123],"identify":[88],"locate":[90],"generalization":[91],"properties":[92],"invisible":[97],"via":[98],"in-distribution":[99],"test":[100],"set":[101,127],"performance,":[102],"(iii)":[104],"new":[105],"evaluations":[106],"how":[108],"features":[111],"vary":[112],"as":[113],"width":[114],"depth":[116],"increased.Our":[118],"results":[119],"suggest":[120],"promising":[126],"tools":[129],"shedding":[131],"light":[132],"on":[133],"models.":[139]},"counts_by_year":[{"year":2026,"cited_by_count":2},{"year":2025,"cited_by_count":1},{"year":2024,"cited_by_count":4}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2023-12-10T00:00:00"}