{"id":"https://openalex.org/W4416374877","doi":"https://doi.org/10.48550/arxiv.2504.15843","title":"Pre-DPO: Improving Data Utilization in Direct Preference Optimization Using a Guiding Reference Model","display_name":"Pre-DPO: Improving Data Utilization in Direct Preference Optimization Using a Guiding Reference Model","publication_year":2025,"publication_date":"2025-04-22","ids":{"openalex":"https://openalex.org/W4416374877","doi":"https://doi.org/10.48550/arxiv.2504.15843"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2504.15843","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2504.15843","pdf_url":"https://arxiv.org/pdf/2504.15843","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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"},"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2504.15843","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Pan, Junshu","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Pan, Junshu","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5013863365","display_name":"Weifeng Shen","orcid":"https://orcid.org/0000-0002-0418-6848"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shen, Wei","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Huang, Shulin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Shulin","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5071444362","display_name":"Qiji Zhou","orcid":"https://orcid.org/0000-0001-5297-4796"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Qiji","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5100333738","display_name":"Yue Zhang","orcid":"https://orcid.org/0000-0002-6327-5023"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Yue","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":5,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.2915000021457672,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.2915000021457672,"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/T10462","display_name":"Reinforcement Learning in Robotics","score":0.1306000053882599,"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/T10203","display_name":"Recommender Systems and Techniques","score":0.0835999995470047,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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/robustness","display_name":"Robustness (evolution)","score":0.7013000249862671},{"id":"https://openalex.org/keywords/preference","display_name":"Preference","score":0.6025000214576721},{"id":"https://openalex.org/keywords/reference-model","display_name":"Reference model","score":0.5519999861717224},{"id":"https://openalex.org/keywords/initialization","display_name":"Initialization","score":0.5489000082015991},{"id":"https://openalex.org/keywords/independent-and-identically-distributed-random-variables","display_name":"Independent and identically distributed random variables","score":0.4050000011920929},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.3393999934196472}],"concepts":[{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.7013000249862671},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6908000111579895},{"id":"https://openalex.org/C2781249084","wikidata":"https://www.wikidata.org/wiki/Q908656","display_name":"Preference","level":2,"score":0.6025000214576721},{"id":"https://openalex.org/C150189527","wikidata":"https://www.wikidata.org/wiki/Q356674","display_name":"Reference model","level":2,"score":0.5519999861717224},{"id":"https://openalex.org/C114466953","wikidata":"https://www.wikidata.org/wiki/Q6034165","display_name":"Initialization","level":2,"score":0.5489000082015991},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.42820000648498535},{"id":"https://openalex.org/C141513077","wikidata":"https://www.wikidata.org/wiki/Q378542","display_name":"Independent and identically distributed random variables","level":3,"score":0.4050000011920929},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.35749998688697815},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.34599998593330383},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.3393999934196472},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.32910001277923584},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.31520000100135803},{"id":"https://openalex.org/C60478076","wikidata":"https://www.wikidata.org/wiki/Q3036835","display_name":"Reference data","level":2,"score":0.3052000105381012},{"id":"https://openalex.org/C137836250","wikidata":"https://www.wikidata.org/wiki/Q984063","display_name":"Optimization problem","level":2,"score":0.28279998898506165},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.26930001378059387},{"id":"https://openalex.org/C2780440489","wikidata":"https://www.wikidata.org/wiki/Q5227278","display_name":"Data-driven","level":2,"score":0.25360000133514404}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2504.15843","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2504.15843","pdf_url":"https://arxiv.org/pdf/2504.15843","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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.2504.15843","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2504.15843","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2504.15843","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2504.15843","pdf_url":"https://arxiv.org/pdf/2504.15843","source":{"id":"https://openalex.org/S4393918464","display_name":"ArXiv.org","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416374877.pdf","grobid_xml":"https://content.openalex.org/works/W4416374877.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Direct":[0],"Preference":[1,77],"Optimization":[2,78],"(DPO)":[3],"simplifies":[4],"reinforcement":[5],"learning":[6],"from":[7],"human":[8,19],"feedback":[9],"(RLHF)":[10],"for":[11,145],"large":[12],"language":[13],"models":[14,53,180],"(LLMs)":[15],"by":[16,109],"directly":[17],"optimizing":[18],"preferences":[20],"without":[21,176],"an":[22],"explicit":[23],"reward":[24],"model.":[25,114],"We":[26],"find":[27],"that":[28,104,136,165],"during":[29],"DPO":[30,56,173],"training,":[31],"the":[32,36,44,49,69,121,127,146,169],"reference":[33,52,73,113,116],"model":[34,74,117,147],"plays":[35],"role":[37],"of":[38,47,71,171],"a":[39,65,72,97,111,133],"data":[40,61],"weight":[41],"adjuster.":[42],"However,":[43],"common":[45],"practice":[46],"initializing":[48],"policy":[50,123],"and":[51,63,83,148,160,174],"identically":[54],"in":[55,75],"can":[57],"lead":[58],"to":[59,87,141,151],"inefficient":[60],"utilization":[62],"impose":[64],"performance":[66,108,170],"ceiling.":[67],"Meanwhile,":[68],"lack":[70],"Simple":[76],"(SimPO)":[79],"reduces":[80],"training":[81,102,128],"robustness":[82],"necessitates":[84],"stricter":[85],"conditions":[86],"prevent":[88],"catastrophic":[89],"forgetting.":[90],"In":[91],"this":[92],"work,":[93],"we":[94],"propose":[95],"Pre-DPO,":[96],"simple":[98],"yet":[99],"effective":[100],"DPO-based":[101],"paradigm":[103],"enhances":[105],"preference":[106,129],"optimization":[107],"leveraging":[110],"guiding":[112,134],"This":[115],"provides":[118],"foresight":[119],"into":[120],"optimal":[122],"state":[124],"achievable":[125],"through":[126],"data,":[130],"serving":[131],"as":[132],"mechanism":[135],"adaptively":[137],"assigns":[138],"higher":[139],"weights":[140,150],"samples":[142],"more":[143],"suitable":[144],"lower":[149],"those":[152],"less":[153],"suitable.":[154],"Extensive":[155],"experiments":[156],"on":[157,178],"AlpacaEval":[158],"2.0":[159],"Arena-Hard":[161],"v0.1":[162],"benchmarks":[163],"demonstrate":[164],"Pre-DPO":[166],"consistently":[167],"improves":[168],"both":[172],"SimPO,":[175],"relying":[177],"external":[179],"or":[181],"additional":[182],"data.":[183]},"counts_by_year":[],"updated_date":"2026-04-21T08:09:41.155169","created_date":"2025-10-10T00:00:00"}