{"id":"https://openalex.org/W7166891695","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1444","title":"GFT: From Imitation to Reward Fine-Tuning with Unbiased Group Advantages and Dynamic Coefficient Rectification","display_name":"GFT: From Imitation to Reward Fine-Tuning with Unbiased Group Advantages and Dynamic Coefficient Rectification","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166891695","doi":"https://doi.org/10.18653/v1/2026.findings-acl.1444"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-acl.1444","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1444","pdf_url":"https://aclanthology.org/2026.findings-acl.1444.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":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-acl.1444.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5122913392","display_name":"Wangjie Gan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wangjie Gan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5132548705","display_name":"Miao Pan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Miao Pan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139760912","display_name":"Linbo Xi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Linbo Xi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139751207","display_name":"Wenqi Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wenqi Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139753128","display_name":"Jintao Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jintao Chen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139776456","display_name":"Jianwei Yin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jianwei Yin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5101722406","display_name":"Xuhong Zhang","orcid":"https://orcid.org/0000-0002-8571-9780"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xuhong Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"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":0.87560115,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"28909","last_page":"28922"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11034","display_name":"Digital Filter Design and Implementation","score":0.29440000653266907,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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/T11034","display_name":"Digital Filter Design and Implementation","score":0.29440000653266907,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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/T11210","display_name":"Mathematical Analysis and Transform Methods","score":0.03519999980926514,"subfield":{"id":"https://openalex.org/subfields/2604","display_name":"Applied Mathematics"},"field":{"id":"https://openalex.org/fields/26","display_name":"Mathematics"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12300","display_name":"Advanced Electrical Measurement Techniques","score":0.025100000202655792,"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/rectification","display_name":"Rectification","score":0.7268000245094299},{"id":"https://openalex.org/keywords/group","display_name":"Group (periodic table)","score":0.5478000044822693},{"id":"https://openalex.org/keywords/imitation","display_name":"Imitation","score":0.48080000281333923},{"id":"https://openalex.org/keywords/control-theory","display_name":"Control theory (sociology)","score":0.3262999951839447},{"id":"https://openalex.org/keywords/matching","display_name":"Matching (statistics)","score":0.2913999855518341}],"concepts":[{"id":"https://openalex.org/C50942859","wikidata":"https://www.wikidata.org/wiki/Q4967193","display_name":"Rectification","level":3,"score":0.7268000245094299},{"id":"https://openalex.org/C2781311116","wikidata":"https://www.wikidata.org/wiki/Q83306","display_name":"Group (periodic table)","level":2,"score":0.5478000044822693},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.49559998512268066},{"id":"https://openalex.org/C126388530","wikidata":"https://www.wikidata.org/wiki/Q1131737","display_name":"Imitation","level":2,"score":0.48080000281333923},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.37880000472068787},{"id":"https://openalex.org/C47446073","wikidata":"https://www.wikidata.org/wiki/Q5165890","display_name":"Control theory (sociology)","level":3,"score":0.3262999951839447},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3208000063896179},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.2913999855518341},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2547999918460846},{"id":"https://openalex.org/C2777742833","wikidata":"https://www.wikidata.org/wiki/Q1964083","display_name":"Reciprocal","level":2,"score":0.22789999842643738}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-acl.1444","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1444","pdf_url":"https://aclanthology.org/2026.findings-acl.1444.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":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-acl.1444","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.1444","pdf_url":"https://aclanthology.org/2026.findings-acl.1444.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":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166891695.pdf","grobid_xml":"https://content.openalex.org/works/W7166891695.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2],"are":[3],"typically":[4],"posttrained":[5],"using":[6],"supervised":[7],"fine-tuning":[8],"(SFT)":[9],"and":[10,52,64,95,104,128],"reinforcement":[11],"learning":[12],"(RL),":[13],"yet":[14],"effectively":[15],"unifying":[16],"efficient":[17,118],"knowledge":[18,119],"injection":[19],"with":[20,46,135],"robust":[21],"generalization":[22],"remains":[23],"challenging.In":[24],"this":[25,68],"work,":[26],"we":[27,70],"provide":[28],"a":[29,39,75],"training-dynamics":[30],"analysis":[31],"showing":[32],"that":[33,79,122,131],"SFT":[34],"can":[35],"be":[36],"interpreted":[37],"as":[38],"special":[40],"case":[41],"of":[42],"policy":[43],"gradient":[44,65],"optimization":[45,115],"an":[47],"extremely":[48],"sparse":[49],"implicit":[50],"reward":[51,102],"unstable":[53],"inverse-probability":[54,111],"weighting,":[55],"which":[56,90,108],"together":[57],"lead":[58],"to":[59,100,113],"single-path":[60],"dependency,":[61],"entropy":[62],"collapse,":[63],"explosion.Motivated":[66],"by":[67],"diagnosis,":[69],"propose":[71],"Group":[72,87],"Fine-Tuning":[73],"(GFT),":[74],"unified":[76],"post-training":[77],"framework":[78],"addresses":[80],"these":[81],"intrinsic":[82],"limitations":[83],"through":[84],"two":[85],"mechanisms:":[86],"Advantage":[88],"Learning,":[89],"constructs":[91],"diverse":[92],"response":[93],"groups":[94],"derives":[96],"normalized":[97],"contrastive":[98],"supervision":[99],"alleviate":[101],"sparsity,":[103],"Dynamic":[105],"Coefficient":[106],"Rectification,":[107],"adaptively":[109],"bounds":[110],"weights":[112],"stabilize":[114],"while":[116],"preserving":[117],"injection.Experiments":[120],"demonstrate":[121],"GFT":[123],"consistently":[124],"surpasses":[125],"SFT-based":[126],"methods":[127],"yields":[129],"policies":[130],"integrate":[132],"more":[133],"smoothly":[134],"subsequent":[136],"RL":[137],"training.":[138]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}