{"id":"https://openalex.org/W7165201621","doi":"https://doi.org/10.48550/arxiv.2606.18307","title":"DRIFT: Refining Instruction Data via On-Policy Data Attribution","display_name":"DRIFT: Refining Instruction Data via On-Policy Data Attribution","publication_year":2026,"publication_date":"2026-06-16","ids":{"openalex":"https://openalex.org/W7165201621","doi":"https://doi.org/10.48550/arxiv.2606.18307"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.18307","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.18307","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.18307","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138881194","display_name":"Zefan Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Zefan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138856208","display_name":"Lincheng Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Lincheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5138831230","display_name":"Tianyu Yu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu, Tianyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5138923070","display_name":"Yuan Yao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yao, Yuan","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":null,"has_fulltext":false,"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/T10028","display_name":"Topic Modeling","score":0.26840001344680786,"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":0.26840001344680786,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.1680999994277954,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.0746999979019165,"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/correctness","display_name":"Correctness","score":0.6722000241279602},{"id":"https://openalex.org/keywords/weighting","display_name":"Weighting","score":0.5339999794960022},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.531499981880188},{"id":"https://openalex.org/keywords/concept-drift","display_name":"Concept drift","score":0.48890000581741333},{"id":"https://openalex.org/keywords/data-set","display_name":"Data set","score":0.44339999556541443},{"id":"https://openalex.org/keywords/trajectory","display_name":"Trajectory","score":0.38679999113082886},{"id":"https://openalex.org/keywords/stability","display_name":"Stability (learning theory)","score":0.35899999737739563},{"id":"https://openalex.org/keywords/authorship-attribution","display_name":"Authorship attribution","score":0.3424000144004822}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7936999797821045},{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.6722000241279602},{"id":"https://openalex.org/C183115368","wikidata":"https://www.wikidata.org/wiki/Q856577","display_name":"Weighting","level":2,"score":0.5339999794960022},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.531499981880188},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.5182999968528748},{"id":"https://openalex.org/C60777511","wikidata":"https://www.wikidata.org/wiki/Q3045002","display_name":"Concept drift","level":3,"score":0.48890000581741333},{"id":"https://openalex.org/C58489278","wikidata":"https://www.wikidata.org/wiki/Q1172284","display_name":"Data set","level":2,"score":0.44339999556541443},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.41609999537467957},{"id":"https://openalex.org/C13662910","wikidata":"https://www.wikidata.org/wiki/Q193139","display_name":"Trajectory","level":2,"score":0.38679999113082886},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3617999851703644},{"id":"https://openalex.org/C112972136","wikidata":"https://www.wikidata.org/wiki/Q7595718","display_name":"Stability (learning theory)","level":2,"score":0.35899999737739563},{"id":"https://openalex.org/C3020202489","wikidata":"https://www.wikidata.org/wiki/Q2032038","display_name":"Authorship attribution","level":2,"score":0.3424000144004822},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.33629998564720154},{"id":"https://openalex.org/C160920958","wikidata":"https://www.wikidata.org/wiki/Q7662746","display_name":"Synthetic data","level":2,"score":0.3253999948501587},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.31619998812675476},{"id":"https://openalex.org/C115901376","wikidata":"https://www.wikidata.org/wiki/Q184199","display_name":"Automation","level":2,"score":0.3147999942302704},{"id":"https://openalex.org/C2776145971","wikidata":"https://www.wikidata.org/wiki/Q30673951","display_name":"Labeled data","level":2,"score":0.3093000054359436},{"id":"https://openalex.org/C189430467","wikidata":"https://www.wikidata.org/wiki/Q7293293","display_name":"Ranking (information retrieval)","level":2,"score":0.3084000051021576},{"id":"https://openalex.org/C110121322","wikidata":"https://www.wikidata.org/wiki/Q865811","display_name":"Distribution (mathematics)","level":2,"score":0.29649999737739563},{"id":"https://openalex.org/C2778143727","wikidata":"https://www.wikidata.org/wiki/Q1820650","display_name":"Readability","level":2,"score":0.2879999876022339},{"id":"https://openalex.org/C64543145","wikidata":"https://www.wikidata.org/wiki/Q162942","display_name":"Intersection (aeronautics)","level":2,"score":0.26339998841285706},{"id":"https://openalex.org/C60478076","wikidata":"https://www.wikidata.org/wiki/Q3036835","display_name":"Reference data","level":2,"score":0.2590000033378601},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2551000118255615}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.18307","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.18307","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.18307","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.18307","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Optimizing":[0],"the":[1,10,35,56,65,132,142,150,170,188,203],"training":[2,25],"data":[3,19,57,75,210],"distribution":[4,58],"for":[5,120,186],"Supervised":[6,121],"Fine-Tuning":[7],"(SFT)":[8],"dictates":[9],"capability":[11,36],"of":[12,63,124,154,178],"Large":[13],"Language":[14],"Models":[15],"(LLMs).":[16],"While":[17],"existing":[18,209],"curation":[20,211],"methods":[21],"excel":[22],"at":[23],"accelerating":[24],"under":[26],"constrained":[27],"budgets,":[28],"they":[29],"are":[30],"less":[31],"suited":[32],"to":[33,45,54,92,181],"elevating":[34],"upper":[37],"bound.":[38],"The":[39],"challenge":[40],"here":[41],"is":[42],"no":[43],"longer":[44],"identify":[46,82],"a":[47,96,105,175],"smaller":[48],"subset":[49],"that":[50,83,199],"preserves":[51],"performance,":[52],"but":[53],"refine":[55],"toward":[59],"instances":[60],"most":[61],"capable":[62],"improving":[64],"final":[66],"model.":[67],"To":[68],"address":[69],"this":[70,89],"problem,":[71],"we":[72],"explore":[73],"instance-level":[74],"attribution":[76],"using":[77],"Influence":[78,118],"Functions":[79,119],"(IF).":[80],"We":[81,111],"standard":[84],"IF":[85],"formulations":[86],"struggle":[87],"in":[88],"setting":[90],"due":[91],"two":[93],"structural":[94],"limitations:":[95],"proximity":[97,144],"gap":[98,145],"caused":[99],"by":[100],"off-policy":[101],"validation":[102,137,179],"targets,":[103,138],"and":[104,146,165,195],"severe":[106],"bias":[107],"towards":[108],"gradient":[109,171],"norm.":[110],"propose":[112],"DRIFT":[113,130,200],"(Data":[114],"Refinement":[115],"via":[116],"On-Policy":[117],"Fine-Tuning).":[122],"Instead":[123],"relying":[125],"on":[126,162,192,206],"external":[127],"reference":[128],"data,":[129],"utilizes":[131],"model's":[133],"on-policy":[134],"rollouts":[135],"as":[136,183],"which":[139],"empirically":[140],"minimizes":[141],"parameter":[143],"better":[147],"aligns":[148],"with":[149],"local":[151],"neighborhood":[152],"assumption":[153],"IF.":[155],"It":[156],"further":[157],"applies":[158],"signed":[159],"weighting":[160],"based":[161],"trajectory":[163],"correctness":[164],"debiases":[166],"influence":[167],"scores":[168],"against":[169],"hacking":[172],"issue,":[173],"allowing":[174],"small":[176],"set":[177],"queries":[180],"act":[182],"reliable":[184],"anchors":[185],"attributing":[187],"full":[189],"dataset.":[190],"Experiments":[191],"7B-parameter":[193],"instruction":[194],"reasoning":[196],"models":[197],"show":[198],"consistently":[201],"raises":[202],"performance":[204],"ceiling":[205],"both,":[207],"outperforming":[208],"baselines.":[212]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-06-19T00:00:00"}