{"id":"https://openalex.org/W7161551745","doi":"https://doi.org/10.48550/arxiv.2605.15220","title":"Always Learning, Always Mixing: Efficient and Simple Data Mixing All The Time","display_name":"Always Learning, Always Mixing: Efficient and Simple Data Mixing All The Time","publication_year":2026,"publication_date":"2026-05-13","ids":{"openalex":"https://openalex.org/W7161551745","doi":"https://doi.org/10.48550/arxiv.2605.15220"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.15220","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.15220","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":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.2605.15220","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5045250992","display_name":"Michael Y. Hu","orcid":"https://orcid.org/0000-0002-3718-7169"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Michael Y.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136374148","display_name":"Apurva Gandhi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gandhi, Apurva","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136437149","display_name":"Kyunghyun Cho","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cho, Kyunghyun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5081824828","display_name":"Tal Linzen","orcid":"https://orcid.org/0000-0003-0435-6912"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Linzen, Tal","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136407089","display_name":"Pratyusha Sharma","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sharma, Pratyusha","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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.7360000014305115,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.7360000014305115,"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/T10028","display_name":"Topic Modeling","score":0.05209999904036522,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.03669999912381172,"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/mixing","display_name":"Mixing (physics)","score":0.6313999891281128},{"id":"https://openalex.org/keywords/retraining","display_name":"Retraining","score":0.5196999907493591},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.51910001039505},{"id":"https://openalex.org/keywords/fraction","display_name":"Fraction (chemistry)","score":0.5006999969482422},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.4553000032901764},{"id":"https://openalex.org/keywords/proxy","display_name":"Proxy (statistics)","score":0.43209999799728394},{"id":"https://openalex.org/keywords/data-modeling","display_name":"Data modeling","score":0.38909998536109924},{"id":"https://openalex.org/keywords/sequence","display_name":"Sequence (biology)","score":0.38909998536109924},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.38280001282691956},{"id":"https://openalex.org/keywords/distillation","display_name":"Distillation","score":0.37070000171661377}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6761000156402588},{"id":"https://openalex.org/C138777275","wikidata":"https://www.wikidata.org/wiki/Q6884054","display_name":"Mixing (physics)","level":2,"score":0.6313999891281128},{"id":"https://openalex.org/C2778712577","wikidata":"https://www.wikidata.org/wiki/Q3505966","display_name":"Retraining","level":2,"score":0.5196999907493591},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.51910001039505},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5012000203132629},{"id":"https://openalex.org/C149629883","wikidata":"https://www.wikidata.org/wiki/Q660926","display_name":"Fraction (chemistry)","level":2,"score":0.5006999969482422},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.46860000491142273},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.4553000032901764},{"id":"https://openalex.org/C2780148112","wikidata":"https://www.wikidata.org/wiki/Q1432581","display_name":"Proxy (statistics)","level":2,"score":0.43209999799728394},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.38909998536109924},{"id":"https://openalex.org/C2778112365","wikidata":"https://www.wikidata.org/wiki/Q3511065","display_name":"Sequence (biology)","level":2,"score":0.38909998536109924},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.38280001282691956},{"id":"https://openalex.org/C204030448","wikidata":"https://www.wikidata.org/wiki/Q101017","display_name":"Distillation","level":2,"score":0.37070000171661377},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.33629998564720154},{"id":"https://openalex.org/C75553542","wikidata":"https://www.wikidata.org/wiki/Q178161","display_name":"A priori and a posteriori","level":2,"score":0.33009999990463257},{"id":"https://openalex.org/C36503486","wikidata":"https://www.wikidata.org/wiki/Q11235244","display_name":"Domain (mathematical analysis)","level":2,"score":0.3221000134944916},{"id":"https://openalex.org/C160920958","wikidata":"https://www.wikidata.org/wiki/Q7662746","display_name":"Synthetic data","level":2,"score":0.319599986076355},{"id":"https://openalex.org/C21080849","wikidata":"https://www.wikidata.org/wiki/Q13611879","display_name":"Data point","level":2,"score":0.30649998784065247},{"id":"https://openalex.org/C138958017","wikidata":"https://www.wikidata.org/wiki/Q190087","display_name":"Data type","level":2,"score":0.3027999997138977},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.29170000553131104},{"id":"https://openalex.org/C133462117","wikidata":"https://www.wikidata.org/wiki/Q4929239","display_name":"Data collection","level":2,"score":0.2915000021457672},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2912999987602234},{"id":"https://openalex.org/C207685749","wikidata":"https://www.wikidata.org/wiki/Q2088941","display_name":"Domain knowledge","level":2,"score":0.28380000591278076},{"id":"https://openalex.org/C115903097","wikidata":"https://www.wikidata.org/wiki/Q7094097","display_name":"Online machine learning","level":3,"score":0.28040000796318054},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.27250000834465027},{"id":"https://openalex.org/C2983448237","wikidata":"https://www.wikidata.org/wiki/Q1078276","display_name":"Language understanding","level":2,"score":0.263700008392334},{"id":"https://openalex.org/C40506919","wikidata":"https://www.wikidata.org/wiki/Q7452469","display_name":"Sequence learning","level":2,"score":0.259799987077713},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.25690001249313354},{"id":"https://openalex.org/C136336373","wikidata":"https://www.wikidata.org/wiki/Q5611303","display_name":"Grouped data","level":2,"score":0.2567000091075897},{"id":"https://openalex.org/C2776145971","wikidata":"https://www.wikidata.org/wiki/Q30673951","display_name":"Labeled data","level":2,"score":0.2565999925136566},{"id":"https://openalex.org/C2778484313","wikidata":"https://www.wikidata.org/wiki/Q1172540","display_name":"Data stream","level":2,"score":0.2517000138759613}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.15220","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.15220","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":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.15220","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.15220","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":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/16","score":0.5106195211410522,"display_name":"Peace, Justice and strong institutions"}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Data":[0],"mixing":[1,47,87,114,195],"decides":[2],"how":[3],"to":[4,65],"combine":[5],"different":[6,226],"sources":[7],"or":[8],"types":[9],"of":[10,29,53,182,185,208,228,235,243],"data":[11,23,46,86,113,131],"and":[12,35,42,76,101,152,169,211,217],"is":[13,25,40,88,128,156],"a":[14,26,57,66,72,103,112,180,225,233,239],"consequential":[15],"problem":[16,94],"throughout":[17,99],"language":[18,121,229],"model":[19,30,122,230],"training.":[20],"In":[21,188],"pretraining,":[22,166,189],"composition":[24],"key":[27],"determinant":[28],"quality;":[31],"in":[32,159,198],"continual":[33,77,167,170,202],"learning":[34,78,163,244],"adaptation,":[36],"it":[37],"governs":[38],"what":[39],"retained":[41],"acquired.":[43],"Yet":[44],"existing":[45],"methods":[48],"address":[49],"only":[50],"one":[51,96],"phase":[52],"this":[54],"lifecycle":[55],"at":[56],"time:":[58],"some":[59],"require":[60],"smaller":[61],"proxy":[62,150],"models":[63,151],"tied":[64],"single":[67,240],"training":[68,100,123,193],"phase,":[69],"others":[70],"assume":[71],"fixed":[73],"domain":[74],"set,":[75],"lacks":[79],"principled":[80],"guidance":[81],"altogether.":[82],"We":[83,107],"argue":[84],"that":[85,97,116,129],"fundamentally":[89],"an":[90],"online":[91],"decision":[92],"making":[93],"--":[95],"recurs":[98],"demands":[102],"single,":[104],"unified":[105],"solution.":[106],"introduce":[108],"OP-Mix":[109,173,190,204,223],"(On-Policy":[110],"Mix),":[111],"algorithm":[115],"operates":[117],"across":[118],"the":[119,145,154,160,183,186,206],"entire":[120],"lifecycle.":[124],"Our":[125],"main":[126],"insight":[127],"candidate":[130],"mixtures":[132,177],"can":[133],"be":[134],"cheaply":[135],"simulated":[136],"by":[137,196],"interpolating":[138],"between":[139],"low-rank":[140],"adapters":[141],"trained":[142],"directly":[143],"on":[144],"current":[146],"model,":[147],"eliminating":[148],"separate":[149],"ensuring":[153],"search":[155],"always":[157],"grounded":[158],"model's":[161],"actual":[162],"dynamics.":[164],"Across":[165],"midtraining,":[168],"instruction":[171],"tuning,":[172],"consistently":[174],"finds":[175],"near-optimal":[176],"while":[178,214],"using":[179,215],"fraction":[181],"compute":[184],"baselines.":[187],"improves":[191],"upon":[192],"without":[194],"6.3%":[197],"average":[199],"perplexity.":[200],"For":[201],"learning,":[203],"matches":[205],"performance":[207],"both":[209],"retraining":[210],"on-policy":[212],"distillation":[213],"66%":[216],"95%":[218],"less":[219],"overall":[220],"compute,":[221],"respectively.":[222],"suggests":[224],"view":[227],"training:":[231],"not":[232],"sequence":[234],"distinct":[236],"phases,":[237],"but":[238],"continuous":[241],"process":[242],"from":[245],"data.":[246]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-19T00:00:00"}