{"id":"https://openalex.org/W7166807961","doi":"https://doi.org/10.18653/v1/2026.acl-long.71","title":"Flip-Flop Consistency: Unsupervised Training for Robustness to Prompt Perturbations in LLMs","display_name":"Flip-Flop Consistency: Unsupervised Training for Robustness to Prompt Perturbations in LLMs","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166807961","doi":"https://doi.org/10.18653/v1/2026.acl-long.71"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.acl-long.71","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.71","pdf_url":"https://aclanthology.org/2026.acl-long.71.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 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.acl-long.71.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5139843660","display_name":"Parsa Hejabi","orcid":null},"institutions":[{"id":"https://openalex.org/I1174212","display_name":"University of Southern California","ror":"https://ror.org/03taz7m60","country_code":"US","type":"education","lineage":["https://openalex.org/I1174212"]},{"id":"https://openalex.org/I2800817003","display_name":"California Southern University","ror":"https://ror.org/058zz0t50","country_code":"US","type":"education","lineage":["https://openalex.org/I2800817003"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Parsa Hejabi","raw_affiliation_strings":["University of Southern California"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Southern California","institution_ids":["https://openalex.org/I1174212","https://openalex.org/I2800817003"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139750620","display_name":"Elnaz Rahmati","orcid":null},"institutions":[{"id":"https://openalex.org/I1174212","display_name":"University of Southern California","ror":"https://ror.org/03taz7m60","country_code":"US","type":"education","lineage":["https://openalex.org/I1174212"]},{"id":"https://openalex.org/I2800817003","display_name":"California Southern University","ror":"https://ror.org/058zz0t50","country_code":"US","type":"education","lineage":["https://openalex.org/I2800817003"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Elnaz Rahmati","raw_affiliation_strings":["University of Southern California"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Southern California","institution_ids":["https://openalex.org/I1174212","https://openalex.org/I2800817003"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5055105286","display_name":"Alireza S. Ziabari","orcid":null},"institutions":[{"id":"https://openalex.org/I1174212","display_name":"University of Southern California","ror":"https://ror.org/03taz7m60","country_code":"US","type":"education","lineage":["https://openalex.org/I1174212"]},{"id":"https://openalex.org/I2800817003","display_name":"California Southern University","ror":"https://ror.org/058zz0t50","country_code":"US","type":"education","lineage":["https://openalex.org/I2800817003"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Alireza Salkhordeh Ziabari","raw_affiliation_strings":["University of Southern California"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Southern California","institution_ids":["https://openalex.org/I1174212","https://openalex.org/I2800817003"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5065952016","display_name":"Morteza Dehghani","orcid":"https://orcid.org/0000-0002-9478-4365"},"institutions":[{"id":"https://openalex.org/I1174212","display_name":"University of Southern California","ror":"https://ror.org/03taz7m60","country_code":"US","type":"education","lineage":["https://openalex.org/I1174212"]},{"id":"https://openalex.org/I2800817003","display_name":"California Southern University","ror":"https://ror.org/058zz0t50","country_code":"US","type":"education","lineage":["https://openalex.org/I2800817003"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Morteza Dehghani","raw_affiliation_strings":["University of Southern California"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Southern California","institution_ids":["https://openalex.org/I1174212","https://openalex.org/I2800817003"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.79795723,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"1571","last_page":"1587"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.18559999763965607,"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.18559999763965607,"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/T10653","display_name":"Robot Manipulation and Learning","score":0.09130000323057175,"subfield":{"id":"https://openalex.org/subfields/2207","display_name":"Control and Systems 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/T11195","display_name":"Simulation Techniques and Applications","score":0.029600000008940697,"subfield":{"id":"https://openalex.org/subfields/1803","display_name":"Management Science and Operations Research"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/robustness","display_name":"Robustness (evolution)","score":0.6510999798774719},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.38280001282691956},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.32440000772476196},{"id":"https://openalex.org/keywords/unsupervised-learning","display_name":"Unsupervised learning","score":0.31700000166893005}],"concepts":[{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.6510999798774719},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5400000214576721},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5077000260353088},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.38280001282691956},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.36890000104904175},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.32440000772476196},{"id":"https://openalex.org/C8038995","wikidata":"https://www.wikidata.org/wiki/Q1152135","display_name":"Unsupervised learning","level":2,"score":0.31700000166893005},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.263700008392334},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2574000060558319},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.25529998540878296}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.acl-long.71","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.71","pdf_url":"https://aclanthology.org/2026.acl-long.71.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 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.acl-long.71","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.71","pdf_url":"https://aclanthology.org/2026.acl-long.71.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 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/13","score":0.7883013486862183,"display_name":"Climate action"}],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166807961.pdf","grobid_xml":"https://content.openalex.org/works/W7166807961.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"often":[4],"produce":[5],"inconsistent":[6],"answers":[7],"when":[8,138],"faced":[9],"with":[10,90],"different":[11],"phrasings":[12],"of":[13,40,144],"the":[14,73],"same":[15],"prompt.In":[16],"this":[17],"paper,":[18],"we":[19],"propose":[20],"Flip-Flop":[21],"Consistency":[22],"(F":[23],"2":[24,36,98,122,153,167],"C),":[25],"an":[26,170],"unsupervised":[27,172],"training":[28],"method":[29,82,173],"that":[30,66],"improves":[31,105,156],"robustness":[32],"to":[33,55],"such":[34],"perturbations.F":[35],"C":[37,99,123,154,168],"is":[38,61],"composed":[39],"two":[41],"key":[42],"components.The":[43],"first,":[44],"Consensus":[45],"Cross-Entropy":[46],"(CCE),":[47],"uses":[48],"a":[49,57,62,142],"majority":[50],"vote":[51],"across":[52,115,134],"prompt":[53,92,145,182],"variations":[54,93],"create":[56],"hard":[58],"pseudo-label.The":[59],"second":[60],"representation":[63],"alignment":[64],"loss":[65],"pulls":[67],"lower-confidence":[68],"and":[69,111,129,147,159,179],"non-majority":[70],"predictors":[71],"toward":[72],"consensus":[74],"established":[75],"by":[76,103,109,117],"high-confidence,":[77],"majorityvoting":[78],"variations.We":[79],"evaluate":[80],"our":[81],"on":[83,140,149],"11":[84],"datasets":[85],"spanning":[86],"four":[87],"NLP":[88],"tasks,":[89],"4-15":[91],"per":[94],"dataset.On":[95],"average,":[96],"F":[97,107,121,127,152,166],"raises":[100],"observed":[101],"agreement":[102,130,160],"11.62%,":[104],"mean":[106],"1":[108,128,184],"8.94%,":[110],"reduces":[112],"performance":[113,158],"variance":[114,133],"formats":[116],"3.29%.In":[118],"out-of-domain":[119],"evaluations,":[120],"generalizes":[124],"effectively,":[125],"increasing":[126],"while":[131,161],"decreasing":[132],"most":[135],"source-target":[136],"pairs.Finally,":[137],"trained":[139],"only":[141],"subset":[143],"perturbations":[146],"evaluated":[148],"held-out":[150],"formats,":[151],"consistently":[155],"both":[157],"reducing":[162],"variance.These":[163],"findings":[164],"highlight":[165],"as":[169],"effective":[171],"for":[174],"enhancing":[175],"LLM":[176],"consistency,":[177],"performance,":[178],"generalization":[180],"under":[181],"perturbations.":[183]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}