{"id":"https://openalex.org/W7138260408","doi":"https://doi.org/10.1609/aaai.v40i38.40469","title":"DRIFT: Difference-Aware Reinforcement Through Iterative Fine-Tuning for Language Model","display_name":"DRIFT: Difference-Aware Reinforcement Through Iterative Fine-Tuning for Language Model","publication_year":2026,"publication_date":"2026-03-14","ids":{"openalex":"https://openalex.org/W7138260408","doi":"https://doi.org/10.1609/aaai.v40i38.40469"},"language":"en","primary_location":{"id":"doi:10.1609/aaai.v40i38.40469","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i38.40469","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://doi.org/10.1609/aaai.v40i38.40469","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5126587577","display_name":"Wenjie Liao","orcid":null},"institutions":[{"id":"https://openalex.org/I180662265","display_name":"China Mobile (China)","ror":"https://ror.org/05gftfe97","country_code":"CN","type":"company","lineage":["https://openalex.org/I180662265"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wenjie Liao","raw_affiliation_strings":["Guangdong OPPO Mobile Telecommunications Corp.,Ltd"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Guangdong OPPO Mobile Telecommunications Corp.,Ltd","institution_ids":["https://openalex.org/I180662265"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129672818","display_name":"Xiaohui Song","orcid":null},"institutions":[{"id":"https://openalex.org/I180662265","display_name":"China Mobile (China)","ror":"https://ror.org/05gftfe97","country_code":"CN","type":"company","lineage":["https://openalex.org/I180662265"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xiaohui Song","raw_affiliation_strings":["Guangdong OPPO Mobile Telecommunications Corp.,Ltd"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Guangdong OPPO Mobile Telecommunications Corp.,Ltd","institution_ids":["https://openalex.org/I180662265"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5129693160","display_name":"Haonan Lu","orcid":null},"institutions":[{"id":"https://openalex.org/I180662265","display_name":"China Mobile (China)","ror":"https://ror.org/05gftfe97","country_code":"CN","type":"company","lineage":["https://openalex.org/I180662265"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Haonan Lu","raw_affiliation_strings":["Guangdong OPPO Mobile Telecommunications Corp.,Ltd"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Guangdong OPPO Mobile Telecommunications Corp.,Ltd","institution_ids":["https://openalex.org/I180662265"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I180662265"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":"40","issue":"38","first_page":"31988","last_page":"31996"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.7540000081062317,"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":0.7540000081062317,"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.06239999830722809,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}},{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.030500000342726707,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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/automatic-summarization","display_name":"Automatic summarization","score":0.7160999774932861},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.675599992275238},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"Reinforcement learning","score":0.6065999865531921},{"id":"https://openalex.org/keywords/vocabulary","display_name":"Vocabulary","score":0.5389999747276306},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.5004000067710876},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.446399986743927},{"id":"https://openalex.org/keywords/concept-drift","display_name":"Concept drift","score":0.4332999885082245},{"id":"https://openalex.org/keywords/train","display_name":"Train","score":0.423799991607666},{"id":"https://openalex.org/keywords/structured-prediction","display_name":"Structured prediction","score":0.39410001039505005}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7849000096321106},{"id":"https://openalex.org/C170858558","wikidata":"https://www.wikidata.org/wiki/Q1394144","display_name":"Automatic summarization","level":2,"score":0.7160999774932861},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.675599992275238},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6065999865531921},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5778999924659729},{"id":"https://openalex.org/C2777601683","wikidata":"https://www.wikidata.org/wiki/Q6499736","display_name":"Vocabulary","level":2,"score":0.5389999747276306},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.5004000067710876},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.446399986743927},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4392000138759613},{"id":"https://openalex.org/C60777511","wikidata":"https://www.wikidata.org/wiki/Q3045002","display_name":"Concept drift","level":3,"score":0.4332999885082245},{"id":"https://openalex.org/C190839683","wikidata":"https://www.wikidata.org/wiki/Q2448197","display_name":"Train","level":2,"score":0.423799991607666},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.39410001039505005},{"id":"https://openalex.org/C117619785","wikidata":"https://www.wikidata.org/wiki/Q6094414","display_name":"Iterative learning control","level":3,"score":0.37619999051094055},{"id":"https://openalex.org/C157657479","wikidata":"https://www.wikidata.org/wiki/Q2367247","display_name":"Closed captioning","level":3,"score":0.3578000068664551},{"id":"https://openalex.org/C183322885","wikidata":"https://www.wikidata.org/wiki/Q17007702","display_name":"Context model","level":3,"score":0.3564999997615814},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.33169999718666077},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3231000006198883},{"id":"https://openalex.org/C2777402240","wikidata":"https://www.wikidata.org/wiki/Q6783436","display_name":"Masking (illustration)","level":2,"score":0.31290000677108765},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.31040000915527344},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.30660000443458557},{"id":"https://openalex.org/C2780980858","wikidata":"https://www.wikidata.org/wiki/Q110022","display_name":"Dual (grammatical number)","level":2,"score":0.28690001368522644},{"id":"https://openalex.org/C2776187449","wikidata":"https://www.wikidata.org/wiki/Q1513879","display_name":"Natural language generation","level":3,"score":0.2833999991416931},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.2800999879837036},{"id":"https://openalex.org/C103088060","wikidata":"https://www.wikidata.org/wiki/Q1062839","display_name":"Error detection and correction","level":2,"score":0.2770000100135803},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.27549999952316284},{"id":"https://openalex.org/C2779439875","wikidata":"https://www.wikidata.org/wiki/Q1078276","display_name":"Natural language understanding","level":3,"score":0.27399998903274536},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.266400009393692},{"id":"https://openalex.org/C49937458","wikidata":"https://www.wikidata.org/wiki/Q2599292","display_name":"Probabilistic logic","level":2,"score":0.25780001282691956},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.2563000023365021},{"id":"https://openalex.org/C45493050","wikidata":"https://www.wikidata.org/wiki/Q7884934","display_name":"Unified Model","level":2,"score":0.2547000050544739},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.250900000333786}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1609/aaai.v40i38.40469","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i38.40469","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},{"id":"pmh:oai:ojs.aaai.org:article/40469","is_oa":false,"landing_page_url":"https://ojs.aaai.org/index.php/AAAI/article/view/40469","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"2159-5399","raw_type":"info:eu-repo/semantics/article"}],"best_oa_location":{"id":"doi:10.1609/aaai.v40i38.40469","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i38.40469","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"sustainable_development_goals":[{"display_name":"Quality Education","score":0.8248083591461182,"id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Self-play":[0],"fine-tuning":[1,165,169],"has":[2],"emerged":[3],"as":[4],"a":[5,70],"promising":[6],"approach":[7],"to":[8,47,55,126,173,199],"improve":[9],"Large":[10],"Language":[11],"Models":[12],"(LLMs)":[13],"without":[14],"additional":[15],"human":[16],"annotations.":[17],"However,":[18],"existing":[19],"methods":[20],"struggle":[21],"with":[22,37],"complex":[23],"generation":[24,215],"tasks":[25,180],"requiring":[26],"long":[27],"context":[28],"understanding,":[29],"where":[30],"models":[31,125],"produce":[32],"partially":[33],"correct":[34,127],"outputs":[35,97],"interleaved":[36],"errors.":[38],"Traditional":[39],"approaches":[40],"train":[41],"on":[42,77,104,148,176,207],"entire":[43],"sequences":[44],"uniformly,":[45],"failing":[46],"distinguish":[48],"between":[49,95],"well-predicted":[50],"and":[51,59,91,98,130,154,167,194],"erroneous":[52],"regions,":[53],"leading":[54],"diluted":[56],"learning":[57],"signals":[58],"slow":[60],"convergence.":[61],"We":[62],"propose":[63],"DRIFT":[64,80,138,147,160,186],"(Difference-aware":[65],"Reinforcement":[66],"through":[67],"Iterative":[68],"Fine-Tuning),":[69],"novel":[71],"self-play":[72,168,201],"framework":[73],"that":[74,89,111,137,159,204],"selectively":[75],"trains":[76],"prediction":[78,208],"differences.":[79],"introduces":[81],"two":[82],"key":[83],"innovations:":[84],"(1)":[85],"Difference-Aware":[86],"Masking":[87],"(DAM)":[88],"identifies":[90],"masks":[92],"common":[93],"subsequences":[94],"model":[96],"ground":[99],"truth,":[100],"focusing":[101],"training":[102],"exclusively":[103],"error":[105],"regions;":[106],"(2)":[107],"Occurrence-Aware":[108],"Loss":[109],"(OAL)":[110],"provides":[112],"position-invariant":[113],"vocabulary":[114],"supervision,":[115],"complementing":[116],"the":[117,188],"position-sensitive":[118],"adversarial":[119],"loss.":[120],"This":[121],"dual":[122],"mechanism":[123],"enables":[124],"both":[128,163],"positional":[129],"lexical":[131],"errors":[132],"effectively.":[133],"Theoretically,":[134],"we":[135,145],"prove":[136],"converges":[139],"when":[140],"masked":[141],"distributions":[142],"align.":[143],"Empirically,":[144],"evaluate":[146],"diverse":[149],"summarization":[150,179],"benchmarks":[151],"using":[152],"Qwen2.5-3B":[153],"LLaMA-3.1-8B":[155],"models.":[156],"Results":[157],"show":[158],"significantly":[161],"outperforms":[162],"supervised":[164],"(SFT)":[166],"(SPIN),":[170],"achieving":[171],"up":[172],"16\\%":[174],"improvement":[175],"SAMSum":[177],"dialogue":[178],"while":[181],"maintaining":[182],"general":[183],"capabilities.":[184],"Notably,":[185],"breaks":[187],"performance":[189],"ceiling":[190],"of":[191],"continued":[192],"SFT":[193],"demonstrates":[195],"superior":[196],"efficiency":[197],"compared":[198],"holistic":[200],"methods,":[202],"validating":[203],"targeted":[205],"optimization":[206],"differences":[209],"is":[210],"crucial":[211],"for":[212],"structured":[213],"text":[214],"tasks.":[216]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2026-03-18T00:00:00"}