{"id":"https://openalex.org/W4416036350","doi":"https://doi.org/10.18653/v1/2025.emnlp-main.776","title":"DyePack: Provably Flagging Test Set Contamination in LLMs Using Backdoors","display_name":"DyePack: Provably Flagging Test Set Contamination in LLMs Using Backdoors","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416036350","doi":"https://doi.org/10.18653/v1/2025.emnlp-main.776"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.emnlp-main.776","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.776","pdf_url":"https://aclanthology.org/2025.emnlp-main.776.pdf","source":null,"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 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.emnlp-main.776.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Yize Cheng","orcid":null},"institutions":[{"id":"https://openalex.org/I66946132","display_name":"University of Maryland, College Park","ror":"https://ror.org/047s2c258","country_code":"US","type":"education","lineage":["https://openalex.org/I66946132"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yize Cheng","raw_affiliation_strings":["University of Maryland , College Park"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Maryland , College Park","institution_ids":["https://openalex.org/I66946132"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101701726","display_name":"Wenxiao Wang","orcid":"https://orcid.org/0000-0001-7328-0267"},"institutions":[{"id":"https://openalex.org/I66946132","display_name":"University of Maryland, College Park","ror":"https://ror.org/047s2c258","country_code":"US","type":"education","lineage":["https://openalex.org/I66946132"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Wenxiao Wang","raw_affiliation_strings":["University of Maryland , College Park"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Maryland , College Park","institution_ids":["https://openalex.org/I66946132"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5077078550","display_name":"Mazda Moayeri","orcid":"https://orcid.org/0009-0006-2842-5352"},"institutions":[{"id":"https://openalex.org/I66946132","display_name":"University of Maryland, College Park","ror":"https://ror.org/047s2c258","country_code":"US","type":"education","lineage":["https://openalex.org/I66946132"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Mazda Moayeri","raw_affiliation_strings":["University of Maryland , College Park"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Maryland , College Park","institution_ids":["https://openalex.org/I66946132"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5025450606","display_name":"Soheil Feizi","orcid":"https://orcid.org/0000-0003-0944-8242"},"institutions":[{"id":"https://openalex.org/I66946132","display_name":"University of Maryland, College Park","ror":"https://ror.org/047s2c258","country_code":"US","type":"education","lineage":["https://openalex.org/I66946132"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Soheil Feizi","raw_affiliation_strings":["University of Maryland , College Park"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Maryland , College Park","institution_ids":["https://openalex.org/I66946132"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I66946132"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.43964711,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"15367","last_page":"15384"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12224","display_name":"Nanofabrication and Lithography Techniques","score":0.0820000022649765,"subfield":{"id":"https://openalex.org/subfields/2204","display_name":"Biomedical Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T12224","display_name":"Nanofabrication and Lithography Techniques","score":0.0820000022649765,"subfield":{"id":"https://openalex.org/subfields/2204","display_name":"Biomedical Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12122","display_name":"Physical Unclonable Functions (PUFs) and Hardware Security","score":0.07850000262260437,"subfield":{"id":"https://openalex.org/subfields/1708","display_name":"Hardware and Architecture"},"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/T14117","display_name":"Integrated Circuits and Semiconductor Failure Analysis","score":0.061799999326467514,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/flagging","display_name":"Flagging","score":0.8934000134468079},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.5432000160217285},{"id":"https://openalex.org/keywords/test","display_name":"Test (biology)","score":0.3984000086784363},{"id":"https://openalex.org/keywords/test-set","display_name":"Test set","score":0.35249999165534973},{"id":"https://openalex.org/keywords/component","display_name":"Component (thermodynamics)","score":0.3034000098705292}],"concepts":[{"id":"https://openalex.org/C2777548347","wikidata":"https://www.wikidata.org/wiki/Q5456937","display_name":"Flagging","level":2,"score":0.8934000134468079},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.5432000160217285},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5406000018119812},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.41370001435279846},{"id":"https://openalex.org/C2777267654","wikidata":"https://www.wikidata.org/wiki/Q3519023","display_name":"Test (biology)","level":2,"score":0.3984000086784363},{"id":"https://openalex.org/C169903167","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Test set","level":2,"score":0.35249999165534973},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.3093000054359436},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.3034000098705292},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.2904999852180481},{"id":"https://openalex.org/C112570922","wikidata":"https://www.wikidata.org/wiki/Q60528603","display_name":"Contamination","level":2,"score":0.2761000096797943},{"id":"https://openalex.org/C64869954","wikidata":"https://www.wikidata.org/wiki/Q1859747","display_name":"False positive paradox","level":2,"score":0.2750000059604645},{"id":"https://openalex.org/C58489278","wikidata":"https://www.wikidata.org/wiki/Q1172284","display_name":"Data set","level":2,"score":0.26409998536109924}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.emnlp-main.776","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.776","pdf_url":"https://aclanthology.org/2025.emnlp-main.776.pdf","source":null,"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 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.emnlp-main.776","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.776","pdf_url":"https://aclanthology.org/2025.emnlp-main.776.pdf","source":null,"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 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G6671297155","display_name":null,"funder_award_id":"CAREER","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G8876996369","display_name":null,"funder_award_id":"N00014","funder_id":"https://openalex.org/F4320337345","funder_display_name":"Office of Naval Research"},{"id":"https://openalex.org/G8945643642","display_name":"Institute for Trustworthy AI in Law and Society (TRAILS)","funder_award_id":"2229885","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320332180","display_name":"Defense Advanced Research Projects Agency","ror":"https://ror.org/02caytj08"},{"id":"https://openalex.org/F4320333591","display_name":"Multidisciplinary University Research Initiative","ror":null},{"id":"https://openalex.org/F4320337345","display_name":"Office of Naval Research","ror":"https://ror.org/00rk2pe57"},{"id":"https://openalex.org/F4320338281","display_name":"Army Research Office","ror":"https://ror.org/05epdh915"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416036350.pdf","grobid_xml":"https://content.openalex.org/works/W4416036350.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Open":[0],"benchmarks":[1],"are":[2],"essential":[3],"for":[4,115],"evaluating":[5],"and":[6,13,132,153,166],"advancing":[7],"large":[8],"language":[9],"models,":[10],"offering":[11],"reproducibility":[12],"transparency.However,":[14],"their":[15,66],"accessibility":[16],"makes":[17],"them":[18],"likely":[19],"targets":[20],"of":[21,57,119,179],"test":[22,42,77],"set":[23],"contamination.In":[24],"this":[25],"work,":[26],"we":[27],"introduce":[28],"DyePack,":[29],"a":[30,87,174],"framework":[31],"that":[32,39,82],"leverages":[33],"backdoor":[34,73],"attacks":[35],"to":[36,49,68,79],"identify":[37],"models":[38,81,125,143,170],"used":[40],"benchmark":[41],"sets":[43],"during":[44],"training,":[45],"without":[46],"requiring":[47],"access":[48],"the":[50,58,76],"loss,":[51],"logits,":[52],"or":[53],"any":[54],"internal":[55],"details":[56],"model.Like":[59],"how":[60],"banks":[61],"mix":[62],"dye":[63],"packs":[64],"with":[65,75,93,144,173],"money":[67],"mark":[69],"robbers,":[70],"DyePack":[71,122],"mixes":[72],"samples":[74],"data":[78],"flag":[80],"trained":[83],"on":[84,123,151,155,171],"it.We":[85],"propose":[86],"principled":[88],"design":[89],"incorporating":[90],"multiple":[91],"backdoors":[92],"stochastic":[94],"targets,":[95],"enabling":[96],"exact":[97],"false":[98,109,176],"positive":[99,177],"rate":[100,178],"(FPR)":[101],"computation":[102],"when":[103],"flagging":[104],"every":[105,116],"model.This":[106],"provably":[107],"prevents":[108],"accusations":[110],"while":[111],"providing":[112],"strong":[113],"evidence":[114],"detected":[117],"case":[118],"contamination.We":[120],"evaluate":[121],"five":[124],"across":[126],"three":[127],"datasets,":[128],"covering":[129],"both":[130],"multiplechoice":[131],"open-ended":[133,160],"generation":[134,161],"tasks.For":[135],"multiple-choice":[136],"questions,":[137],"it":[138,163],"successfully":[139],"detects":[140],"all":[141,168],"contaminated":[142,169],"guaranteed":[145,175],"FPRs":[146],"as":[147,149],"low":[148],"0.000073%":[150],"MMLU-Pro":[152],"0.000017%":[154],"Big-Bench-Hard":[156],"using":[157,182],"eight":[158],"backdoors.For":[159],"tasks,":[162],"generalizes":[164],"well":[165],"identifies":[167],"Alpaca":[172],"just":[180],"0.127%":[181],"six":[183],"backdoors.":[184]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2025-11-08T00:00:00"}