{"id":"https://openalex.org/W4399364986","doi":"https://doi.org/10.1145/3630106.3659042","title":"Analyzing And Editing Inner Mechanisms of Backdoored Language Models","display_name":"Analyzing And Editing Inner Mechanisms of Backdoored Language Models","publication_year":2024,"publication_date":"2024-06-03","ids":{"openalex":"https://openalex.org/W4399364986","doi":"https://doi.org/10.1145/3630106.3659042"},"language":"en","primary_location":{"id":"doi:10.1145/3630106.3659042","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3630106.3659042","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"The 2024 ACM Conference on Fairness Accountability and Transparency","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5090231346","display_name":"Max Lamparth","orcid":"https://orcid.org/0000-0001-6405-513X"},"institutions":[{"id":"https://openalex.org/I97018004","display_name":"Stanford University","ror":"https://ror.org/00f54p054","country_code":"US","type":"education","lineage":["https://openalex.org/I97018004"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Max Lamparth","raw_affiliation_strings":["Stanford University, USA"],"raw_orcid":"https://orcid.org/0000-0001-6405-513X","affiliations":[{"raw_affiliation_string":"Stanford University, USA","institution_ids":["https://openalex.org/I97018004"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5059838123","display_name":"Anka Reuel","orcid":"https://orcid.org/0000-0002-7913-9296"},"institutions":[{"id":"https://openalex.org/I97018004","display_name":"Stanford University","ror":"https://ror.org/00f54p054","country_code":"US","type":"education","lineage":["https://openalex.org/I97018004"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Anka Reuel","raw_affiliation_strings":["Stanford University, USA"],"raw_orcid":"https://orcid.org/0000-0002-7913-9296","affiliations":[{"raw_affiliation_string":"Stanford University, USA","institution_ids":["https://openalex.org/I97018004"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I97018004"],"apc_list":null,"apc_paid":null,"fwci":0.3314,"has_fulltext":false,"cited_by_count":2,"citation_normalized_percentile":{"value":0.48311363,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":89,"max":95},"biblio":{"volume":null,"issue":null,"first_page":"2362","last_page":"2373"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9901000261306763,"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/T10028","display_name":"Topic Modeling","score":0.9901000261306763,"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/T10260","display_name":"Software Engineering Research","score":0.9848999977111816,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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":0.9825999736785889,"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/backdoor","display_name":"Backdoor","score":0.9888508915901184},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.7923598885536194},{"id":"https://openalex.org/keywords/robustness","display_name":"Robustness (evolution)","score":0.6189709305763245},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.5938209295272827},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.5628786683082581},{"id":"https://openalex.org/keywords/embedding","display_name":"Embedding","score":0.5159788727760315},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.4906286299228668},{"id":"https://openalex.org/keywords/modeling-language","display_name":"Modeling language","score":0.42316168546676636},{"id":"https://openalex.org/keywords/natural-language-processing","display_name":"Natural language processing","score":0.3863624930381775},{"id":"https://openalex.org/keywords/programming-language","display_name":"Programming language","score":0.22561031579971313},{"id":"https://openalex.org/keywords/computer-security","display_name":"Computer security","score":0.15538910031318665},{"id":"https://openalex.org/keywords/engineering","display_name":"Engineering","score":0.1410999894142151}],"concepts":[{"id":"https://openalex.org/C2781045450","wikidata":"https://www.wikidata.org/wiki/Q254569","display_name":"Backdoor","level":2,"score":0.9888508915901184},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7923598885536194},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.6189709305763245},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.5938209295272827},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.5628786683082581},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.5159788727760315},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4906286299228668},{"id":"https://openalex.org/C179603123","wikidata":"https://www.wikidata.org/wiki/Q1941921","display_name":"Modeling language","level":3,"score":0.42316168546676636},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3863624930381775},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.22561031579971313},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.15538910031318665},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.1410999894142151},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.0},{"id":"https://openalex.org/C185592680","wikidata":"https://www.wikidata.org/wiki/Q2329","display_name":"Chemistry","level":0,"score":0.0},{"id":"https://openalex.org/C55493867","wikidata":"https://www.wikidata.org/wiki/Q7094","display_name":"Biochemistry","level":1,"score":0.0},{"id":"https://openalex.org/C104317684","wikidata":"https://www.wikidata.org/wiki/Q7187","display_name":"Gene","level":2,"score":0.0},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.0},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1145/3630106.3659042","is_oa":false,"landing_page_url":"https://doi.org/10.1145/3630106.3659042","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"The 2024 ACM Conference on Fairness Accountability and Transparency","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":18,"referenced_works":["https://openalex.org/W1490960179","https://openalex.org/W1566289585","https://openalex.org/W2160660844","https://openalex.org/W2612560781","https://openalex.org/W2769358515","https://openalex.org/W2892181857","https://openalex.org/W2930957955","https://openalex.org/W2949676527","https://openalex.org/W2954275542","https://openalex.org/W2970172244","https://openalex.org/W2997591727","https://openalex.org/W3010694149","https://openalex.org/W3042368254","https://openalex.org/W4206070857","https://openalex.org/W4226395669","https://openalex.org/W4236456565","https://openalex.org/W4312768002","https://openalex.org/W4396817314"],"related_works":["https://openalex.org/W4288365749","https://openalex.org/W2936497627","https://openalex.org/W3013624417","https://openalex.org/W4287826556","https://openalex.org/W3098382480","https://openalex.org/W4287598411","https://openalex.org/W3094871513","https://openalex.org/W3100913109","https://openalex.org/W3198458223","https://openalex.org/W3126642501"],"abstract_inverted_index":{"Poisoning":[0],"of":[1,21,25,54,124,149],"data":[2,163],"sets":[3],"is":[4],"a":[5],"potential":[6],"security":[7],"threat":[8],"to":[9,16,38,43,82,98],"large":[10,150],"language":[11,27,57,151],"models":[12,28,58,152],"that":[13,92,142],"can":[14,144],"lead":[15],"backdoored":[17,26,56,132,134],"models.":[18,139],"A":[19],"description":[20],"the":[22,51,68,74,94,101,121,146],"internal":[23,52],"mechanisms":[24,88],"and":[29,59,85,136],"how":[30],"they":[31],"process":[32],"trigger":[33],"inputs,":[34],"e.g.,":[35],"when":[36],"switching":[37],"toxic":[39],"language,":[40],"has":[41],"yet":[42],"be":[44],"found.":[45],"In":[46],"this":[47,80,105],"work,":[48],"we":[49,107,112,143],"study":[50],"representations":[53],"transformer-based":[55],"determine":[60],"early-layer":[61],"MLP":[62,95],"modules":[63,115,157],"as":[64],"most":[65],"important":[66],"for":[67,100],"backdoor":[69,87,102,147],"mechanism":[70],"in":[71],"combination":[72],"with":[73,89,116],"initial":[75],"embedding":[76],"projection.":[77],"We":[78,127,140],"use":[79],"knowledge":[81],"remove,":[83],"insert,":[84],"modify":[86],"engineered":[90],"replacements":[91],"reduce":[93],"module":[96],"outputs":[97],"essentials":[99],"mechanism.":[103],"To":[104],"end,":[106],"introduce":[108],"PCP":[109],"ablation,":[110],"where":[111],"replace":[113],"transformer":[114],"low-rank":[117],"matrices":[118],"based":[119],"on":[120,131,160],"principal":[122],"components":[123],"their":[125],"activations.":[126],"demonstrate":[128],"our":[129],"results":[130],"toy,":[133],"large,":[135],"non-backdoored":[137],"open-source":[138],"show":[141],"improve":[145],"robustness":[148],"by":[153],"locally":[154],"constraining":[155],"individual":[156],"during":[158],"fine-tuning":[159],"potentially":[161],"poisonous":[162],"sets.":[164],"Trigger":[165],"warning:":[166],"Offensive":[167],"language.":[168]},"counts_by_year":[{"year":2025,"cited_by_count":1},{"year":2023,"cited_by_count":1}],"updated_date":"2026-08-18T07:49:30.821534","created_date":"2025-10-10T00:00:00"}