{"id":"https://openalex.org/W7161007357","doi":"https://doi.org/10.48550/arxiv.2605.10981","title":"$\u03be$-DPO: Direct Preference Optimization via Ratio Reward Margin","display_name":"$\u03be$-DPO: Direct Preference Optimization via Ratio Reward Margin","publication_year":2026,"publication_date":"2026-05-09","ids":{"openalex":"https://openalex.org/W7161007357","doi":"https://doi.org/10.48550/arxiv.2605.10981"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2605.10981","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2605.10981","pdf_url":"https://arxiv.org/pdf/2605.10981","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2605.10981","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5136050123","display_name":"Zhengyuan Fan","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I17145004","display_name":"Northwestern Polytechnical University","ror":"https://ror.org/01y0j0j86","country_code":"CN","type":"education","lineage":["https://openalex.org/I17145004"]},{"id":"https://openalex.org/I99418890","display_name":"San Francisco Bay University","ror":"https://ror.org/05wn69s11","country_code":"US","type":"education","lineage":["https://openalex.org/I99418890"]}],"countries":["CN","US"],"is_corresponding":false,"raw_author_name":"Fan, Zhengyuan","raw_affiliation_strings":["School of Computer Science , Northwestern Polytechnical University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computer Science , Northwestern Polytechnical University","institution_ids":["https://openalex.org/I17145004","https://openalex.org/I99418890"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136034301","display_name":"Zhonghua Wu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I17145004","display_name":"Northwestern Polytechnical University","ror":"https://ror.org/01y0j0j86","country_code":"CN","type":"education","lineage":["https://openalex.org/I17145004"]},{"id":"https://openalex.org/I99418890","display_name":"San Francisco Bay University","ror":"https://ror.org/05wn69s11","country_code":"US","type":"education","lineage":["https://openalex.org/I99418890"]}],"countries":["CN","US"],"is_corresponding":false,"raw_author_name":"Wu, Zhonghua","raw_affiliation_strings":["School of Computer Science , Northwestern Polytechnical University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computer Science , Northwestern Polytechnical University","institution_ids":["https://openalex.org/I17145004","https://openalex.org/I99418890"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136070900","display_name":"Yuxuan Du","orcid":"https://orcid.org/0000-0002-1193-9756","observed_orcids":["https://orcid.org/0000-0002-1193-9756"]},"institutions":[{"id":"https://openalex.org/I17145004","display_name":"Northwestern Polytechnical University","ror":"https://ror.org/01y0j0j86","country_code":"CN","type":"education","lineage":["https://openalex.org/I17145004"]},{"id":"https://openalex.org/I99418890","display_name":"San Francisco Bay University","ror":"https://ror.org/05wn69s11","country_code":"US","type":"education","lineage":["https://openalex.org/I99418890"]}],"countries":["CN","US"],"is_corresponding":false,"raw_author_name":"Du, Yuxuan","raw_affiliation_strings":["School of Computer Science , Northwestern Polytechnical University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computer Science , Northwestern Polytechnical University","institution_ids":["https://openalex.org/I17145004","https://openalex.org/I99418890"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5136060841","display_name":"Qun Chen","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I17145004","display_name":"Northwestern Polytechnical University","ror":"https://ror.org/01y0j0j86","country_code":"CN","type":"education","lineage":["https://openalex.org/I17145004"]},{"id":"https://openalex.org/I99418890","display_name":"San Francisco Bay University","ror":"https://ror.org/05wn69s11","country_code":"US","type":"education","lineage":["https://openalex.org/I99418890"]}],"countries":["CN","US"],"is_corresponding":false,"raw_author_name":"Chen, Qun","raw_affiliation_strings":["School of Computer Science , Northwestern Polytechnical University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"School of Computer Science , Northwestern Polytechnical University","institution_ids":["https://openalex.org/I17145004","https://openalex.org/I99418890"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.7178000211715698,"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":0.7178000211715698,"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/T12072","display_name":"Machine Learning and Algorithms","score":0.044599998742341995,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.027699999511241913,"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/preference-optimization","display_name":"preference optimization","score":0.8410000205039978},{"id":"https://openalex.org/keywords/hyperparameter-optimization","display_name":"hyperparameter optimization","score":0.7950000166893005},{"id":"https://openalex.org/keywords/reinforcement-learning-from-human-feedback","display_name":"reinforcement learning from human feedback","score":0.7129999995231628},{"id":"https://openalex.org/keywords/llm-fine-tuning","display_name":"LLM fine-tuning","score":0.6349999904632568}],"concepts":[{"id":"https://openalex.org/C774472","wikidata":"https://www.wikidata.org/wiki/Q6760393","display_name":"Margin (machine learning)","level":2,"score":0.7822999954223633},{"id":"https://openalex.org/C2781249084","wikidata":"https://www.wikidata.org/wiki/Q908656","display_name":"Preference","level":2,"score":0.7010999917984009},{"id":"https://openalex.org/C8642999","wikidata":"https://www.wikidata.org/wiki/Q4171168","display_name":"Hyperparameter","level":2,"score":0.6075000166893005},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.47940000891685486},{"id":"https://openalex.org/C34388435","wikidata":"https://www.wikidata.org/wiki/Q2267362","display_name":"Bounded function","level":2,"score":0.4724000096321106},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.46709999442100525},{"id":"https://openalex.org/C181204326","wikidata":"https://www.wikidata.org/wiki/Q7239820","display_name":"Preference learning","level":3,"score":0.439300000667572},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.41990000009536743},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.4016000032424927},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4000000059604645},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.3734000027179718},{"id":"https://openalex.org/C198531522","wikidata":"https://www.wikidata.org/wiki/Q485146","display_name":"Sample (material)","level":2,"score":0.36959999799728394},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.3483999967575073},{"id":"https://openalex.org/C18555067","wikidata":"https://www.wikidata.org/wiki/Q8375051","display_name":"Joint (building)","level":2,"score":0.319599986076355},{"id":"https://openalex.org/C129848803","wikidata":"https://www.wikidata.org/wiki/Q2564360","display_name":"Sample size determination","level":2,"score":0.3174000084400177},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.3158000111579895},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.2840999960899353},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.27880001068115234},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2728999853134155},{"id":"https://openalex.org/C32172795","wikidata":"https://www.wikidata.org/wiki/Q4692266","display_name":"Aggregation problem","level":2,"score":0.2696000039577484},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.2599000036716461}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2605.10981","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2605.10981","pdf_url":"https://arxiv.org/pdf/2605.10981","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2605.10981","doi":"https://doi.org/10.48550/arxiv.2605.10981","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.10981","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2605.10981","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2605.10981","pdf_url":"https://arxiv.org/pdf/2605.10981","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7161007357.pdf","grobid_xml":"https://content.openalex.org/works/W7161007357.grobid-xml"},"referenced_works_count":3,"referenced_works":["https://openalex.org/W2736601468","https://openalex.org/W2973379954","https://openalex.org/W7196938767"],"related_works":["https://openalex.org/W4416035036","https://openalex.org/W7215107880","https://openalex.org/W4414976198","https://openalex.org/W7164531872","https://openalex.org/W7221009416","https://openalex.org/W7204541737","https://openalex.org/W7160321001","https://openalex.org/W4414586725","https://openalex.org/W4403322564","https://openalex.org/W7137888806"],"abstract_inverted_index":{"Reference-free":[0],"preference":[1,134,144,244],"optimization":[2,135,152,245],"has":[3],"emerged":[4],"as":[5,91],"an":[6,147],"efficient":[7],"alternative":[8],"to":[9,98,161,232],"reinforcement":[10],"learning":[11],"from":[12,154,223],"human":[13],"feedback,":[14],"with":[15,66,115],"Simple":[16],"Preference":[17],"Optimization(SimPO)":[18],"demonstrating":[19],"strong":[20],"performance":[21],"by":[22,108,127,210],"eliminating":[23],"the":[24,33,37,55,103,118,143,151,156,163,174,181,188,203,213,224],"explicit":[25],"reference":[26],"model":[27],"through":[28,146],"a":[29,45,78,92,177,194],"simple":[30],"objective.":[31],"However,":[32],"joint":[34],"tuning":[35],"of":[36,81,158,190],"hyperparameters":[38],"$\u03b2$":[39,191,214],"and":[40,83,123,168,183,192,196,207,215],"$\u03b3$":[41,106,216],"in":[42,58,176,217],"SimPO":[43,59,82,109],"remains":[44],"central":[46],"challenge.":[47],"We":[48,140],"argue":[49],"that":[50,85,239],"this":[51,74],"difficulty":[52],"arises":[53],"because":[54],"margin":[56,105,200,206],"formulation":[57],"is":[60,110,201,208],"not":[61,111],"easily":[62],"interpretable":[63,197],"across":[64,247],"datasets":[65],"different":[67],"reward":[68,93,104,119,138,159,166,175,205,226],"gap":[69,120,227],"structures.":[70],"To":[71],"better":[72],"understand":[73],"issue,":[75],"we":[76,130,172],"conduct":[77],"comprehensive":[79],"analysis":[80],"find":[84],"$\u03b2$,":[86],"rather":[87],"than":[88],"merely":[89],"serving":[90],"scaling":[94],"factor,":[95],"primarily":[96],"functions":[97],"filter":[99],"noisy":[100],"samples,":[101],"while":[102],"defined":[107],"semantically":[112],"well":[113],"aligned":[114],"explicitly":[116],"enlarging":[117],"between":[121,165,180],"chosen":[122,182],"rejected":[124],"responses.":[125],"Motivated":[126],"these":[128],"observations,":[129],"propose":[131],"$\u03be$-DPO:":[132],"Direct":[133],"via":[136],"ratio":[137,178,204],"margin.":[139,198],"first":[141],"reformulate":[142],"objective":[145],"equivalent":[148],"transformation,":[149],"changing":[150],"target":[153],"maximizing":[155],"likelihood":[157],"gaps":[160,167],"minimizing":[162],"distance":[164],"optimal":[169],"margins.":[170],"Then,":[171],"redefine":[173],"form":[179],"rejected,":[184],"which":[185],"effectively":[186],"cancels":[187],"effect":[189],"yields":[193],"bounded":[195],"This":[199],"called":[202],"denoted":[209],"$\u03be$.":[211],"Unlike":[212],"SimPO,":[218],"$\u03be$":[219],"can":[220],"be":[221],"determined":[222],"initial":[225],"distribution,":[228],"thus":[229],"much":[230],"easier":[231],"tune.":[233],"Our":[234],"extensive":[235],"experimental":[236],"results":[237],"show":[238],"$\u03be$-DPO":[240],"substantially":[241],"outperforms":[242],"existing":[243],"methods":[246],"open":[248],"benchmarks":[249],"on":[250],"multiple":[251],"evaluation":[252],"metrics.":[253]},"counts_by_year":[],"updated_date":"2026-11-08T12:12:28.789561","created_date":"2026-05-14T00:00:00"}