{"id":"https://openalex.org/W7224127347","doi":"https://doi.org/10.48550/arxiv.2610.37390","title":"LinMix: Data Mixing as Linear Domain Attribution","display_name":"LinMix: Data Mixing as Linear Domain Attribution","publication_year":2026,"publication_date":"2026-10-27","ids":{"openalex":"https://openalex.org/W7224127347","doi":"https://doi.org/10.48550/arxiv.2610.37390"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.37390","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.37390","pdf_url":"https://arxiv.org/pdf/2610.37390","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":"public-domain","license_id":"https://openalex.org/licenses/public-domain","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/2610.37390","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5059946484","display_name":"Hubert Baniecki","orcid":"https://orcid.org/0000-0001-6661-5364","observed_orcids":["https://orcid.org/0000-0001-6661-5364"]},"institutions":[{"id":"https://openalex.org/I108403487","display_name":"Warsaw University of Technology","ror":"https://ror.org/00y0xnp53","country_code":"PL","type":"education","lineage":["https://openalex.org/I108403487"]},{"id":"https://openalex.org/I4654613","display_name":"University of Warsaw","ror":"https://ror.org/039bjqg32","country_code":"PL","type":"education","lineage":["https://openalex.org/I4654613"]}],"countries":["PL"],"is_corresponding":false,"raw_author_name":"Baniecki, Hubert","raw_affiliation_strings":["Centre for Credible AI , Warsaw University of Technology","University of Warsaw"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Centre for Credible AI , Warsaw University of Technology","institution_ids":["https://openalex.org/I108403487"]},{"raw_affiliation_string":"University of Warsaw","institution_ids":["https://openalex.org/I4654613"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5049061860","display_name":"Przemys\u0142aw Biecek","orcid":"https://orcid.org/0000-0001-8423-1823","observed_orcids":["https://orcid.org/0000-0001-8423-1823"]},"institutions":[{"id":"https://openalex.org/I108403487","display_name":"Warsaw University of Technology","ror":"https://ror.org/00y0xnp53","country_code":"PL","type":"education","lineage":["https://openalex.org/I108403487"]},{"id":"https://openalex.org/I4654613","display_name":"University of Warsaw","ror":"https://ror.org/039bjqg32","country_code":"PL","type":"education","lineage":["https://openalex.org/I4654613"]}],"countries":["PL"],"is_corresponding":false,"raw_author_name":"Biecek, Przemyslaw","raw_affiliation_strings":["Centre for Credible AI , Warsaw University of Technology","University of Warsaw"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Centre for Credible AI , Warsaw University of Technology","institution_ids":["https://openalex.org/I108403487"]},{"raw_affiliation_string":"University of Warsaw","institution_ids":["https://openalex.org/I4654613"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5044448837","display_name":"Martin Pawelczyk","orcid":"https://orcid.org/0000-0002-6191-4434","observed_orcids":["https://orcid.org/0000-0002-6191-4434"]},"institutions":[{"id":"https://openalex.org/I129774422","display_name":"University of Vienna","ror":"https://ror.org/03prydq77","country_code":"AT","type":"education","lineage":["https://openalex.org/I129774422"]}],"countries":["AT"],"is_corresponding":false,"raw_author_name":"Pawelczyk, Martin","raw_affiliation_strings":["Faculty of Computer Science , University of Vienna"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Faculty of Computer Science , University of Vienna","institution_ids":["https://openalex.org/I129774422"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":3,"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/T10028","display_name":"Topic Modeling","score":0.6606000065803528,"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":0.6606000065803528,"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":0.21629999577999115,"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":0.04399999976158142,"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/data-mixing","display_name":"data mixing","score":0.8679999709129333},{"id":"https://openalex.org/keywords/language-model-pretraining","display_name":"language model pretraining","score":0.6919999718666077},{"id":"https://openalex.org/keywords/optimal-experimental-design","display_name":"optimal experimental design","score":0.8330000042915344},{"id":"https://openalex.org/keywords/qwen3-5","display_name":"Qwen3.5","score":0.9369999766349792},{"id":"https://openalex.org/keywords/code-generation","display_name":"code generation","score":0.4659999907016754},{"id":"https://openalex.org/keywords/mathematical-reasoning","display_name":"mathematical reasoning","score":0.7360000014305115}],"concepts":[{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.612500011920929},{"id":"https://openalex.org/C48921125","wikidata":"https://www.wikidata.org/wiki/Q10861030","display_name":"Linear regression","level":2,"score":0.46619999408721924},{"id":"https://openalex.org/C80899671","wikidata":"https://www.wikidata.org/wiki/Q1304193","display_name":"Vertex (graph theory)","level":3,"score":0.4661000072956085},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.44200000166893005},{"id":"https://openalex.org/C83546350","wikidata":"https://www.wikidata.org/wiki/Q1139051","display_name":"Regression","level":2,"score":0.42239999771118164},{"id":"https://openalex.org/C163175372","wikidata":"https://www.wikidata.org/wiki/Q3339222","display_name":"Linear model","level":2,"score":0.4223000109195709},{"id":"https://openalex.org/C36503486","wikidata":"https://www.wikidata.org/wiki/Q11235244","display_name":"Domain (mathematical analysis)","level":2,"score":0.39570000767707825},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.39500001072883606},{"id":"https://openalex.org/C169214877","wikidata":"https://www.wikidata.org/wiki/Q981016","display_name":"Dirichlet distribution","level":3,"score":0.3903000056743622},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.38600000739097595},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.3662000000476837},{"id":"https://openalex.org/C41045048","wikidata":"https://www.wikidata.org/wiki/Q202843","display_name":"Linear programming","level":2,"score":0.3619000017642975},{"id":"https://openalex.org/C62438384","wikidata":"https://www.wikidata.org/wiki/Q331350","display_name":"Simplex","level":2,"score":0.3540000021457672},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.3452000021934509},{"id":"https://openalex.org/C152877465","wikidata":"https://www.wikidata.org/wiki/Q208042","display_name":"Regression analysis","level":2,"score":0.3287999927997589},{"id":"https://openalex.org/C138777275","wikidata":"https://www.wikidata.org/wiki/Q6884054","display_name":"Mixing (physics)","level":2,"score":0.32359999418258667},{"id":"https://openalex.org/C132525143","wikidata":"https://www.wikidata.org/wiki/Q141488","display_name":"Graph","level":2,"score":0.3068000078201294},{"id":"https://openalex.org/C2780148112","wikidata":"https://www.wikidata.org/wiki/Q1432581","display_name":"Proxy (statistics)","level":2,"score":0.301800012588501},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2856000065803528},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.28040000796318054},{"id":"https://openalex.org/C158622935","wikidata":"https://www.wikidata.org/wiki/Q660848","display_name":"Nonlinear system","level":2,"score":0.2784000039100647},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.2653999924659729},{"id":"https://openalex.org/C139945424","wikidata":"https://www.wikidata.org/wiki/Q1940696","display_name":"Mean squared error","level":2,"score":0.26089999079704285}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.37390","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.37390","pdf_url":"https://arxiv.org/pdf/2610.37390","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":"public-domain","license_id":"https://openalex.org/licenses/public-domain","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2610.37390","doi":"https://doi.org/10.48550/arxiv.2610.37390","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.37390","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":"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2610.37390","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.37390","pdf_url":"https://arxiv.org/pdf/2610.37390","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":"public-domain","license_id":"https://openalex.org/licenses/public-domain","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G442103306","display_name":null,"funder_award_id":"FENG.02","funder_id":"https://openalex.org/F4320321042","funder_display_name":"Fundacja na rzecz Nauki Polskiej"}],"funders":[{"id":"https://openalex.org/F4320321042","display_name":"Fundacja na rzecz Nauki Polskiej","ror":"https://ror.org/048zd9m77"},{"id":"https://openalex.org/F4320326815","display_name":"Infrastruktura PL-Grid","ror":"https://ror.org/01m3qaz74"},{"id":"https://openalex.org/F4320337160","display_name":"Academic Computer Centre Cyfronet, AGH University of Science and Technology","ror":"https://ror.org/0575jwx13"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7224127347.pdf","grobid_xml":"https://content.openalex.org/works/W7224127347.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7140092850","https://openalex.org/W7128793074","https://openalex.org/W7140275045","https://openalex.org/W7204266134","https://openalex.org/W7133216852","https://openalex.org/W4400342789","https://openalex.org/W4416034901","https://openalex.org/W4399061634","https://openalex.org/W4414858766","https://openalex.org/W4404390131"],"abstract_inverted_index":{"Language":[0],"model":[1,37,133],"performance":[2,40],"depends":[3],"strongly":[4],"on":[5,29,94,215,219,239],"the":[6,11,42,54,61,67,106,109,137,164,175,183,242],"mixture":[7],"of":[8,71,91,108,166,213],"domains":[9,149],"in":[10],"pretraining":[12],"data":[13,16,73,185,202],"corpus.":[14],"Popular":[15],"mixing":[17,203],"methods":[18],"select":[19],"these":[20],"domain":[21,50,115,127,138],"weights":[22],"by":[23,144,237],"training":[24],"many":[25],"small":[26],"proxy":[27,72],"models":[28,196],"random":[30],"mixes":[31,74,104,180],"and":[32,83,154,169,192,207,217,232],"then":[33],"fitting":[34],"a":[35,49,76,86,229],"regression":[36,55,78],"to":[38,60],"predict":[39],"from":[41],"mix.":[43,186],"We":[44,161],"recognize":[45],"this":[46,65],"practice":[47],"as":[48],"attribution":[51,152,176],"problem,":[52],"where":[53],"measures":[56],"each":[57,114],"domain's":[58],"contribution":[59],"target":[62],"metric.":[63],"Through":[64],"lens,":[66],"standard":[68],"Dirichlet":[69,172],"sampling":[70,125,173,227],"is":[75,116],"poor":[77],"design":[79],"that":[80,113,171,198],"requires":[81,155],"tuning":[82],"cannot":[84],"support":[85],"linear":[87,132,230],"fit":[88],"for":[89],"hundreds":[90],"domains.":[92],"Building":[93],"classical":[95],"optimal":[96,141],"experimental":[97],"design,":[98],"we":[99],"introduce":[100],"LinMix,":[101],"which":[102],"samples":[103],"at":[105,117],"vertices":[107],"capped":[110],"simplex":[111],"so":[112],"its":[118,179],"cap":[119],"or":[120],"absent.":[121],"This":[122],"hyperparameter-free":[123],"vertex":[124,167,226],"decorrelates":[126],"effects,":[128],"enabling":[129],"an":[130],"interpretable":[131],"whose":[134],"coefficients":[135],"are":[136],"attributions.":[139],"The":[140],"mix":[142],"given":[143],"LinMix":[145,199,233],"provably":[146],"keeps":[147],"only":[148,225],"with":[150,189,210],"positive":[151],"values":[153],"no":[156],"search":[157],"over":[158,241],"candidate":[159],"mixes.":[160],"further":[162],"establish":[163],"optimality":[165],"sampling,":[168],"prove":[170],"inflates":[174],"error":[177],"when":[178],"concentrate":[181],"around":[182],"prior":[184],"Pretraining":[187],"experiments":[188],"1B-parameter":[190],"OLMo2":[191],"4B":[193],"Qwen3.5":[194],"language":[195],"show":[197],"outperforms":[200],"Dirichlet-based":[201],"across":[204],"reasoning,":[205],"code,":[206],"math":[208],"benchmarks,":[209],"relative":[211],"gains":[212],"46%":[214],"MBPP":[216],"14%":[218],"GSM8K.":[220],"At":[221],"288":[222],"fine-grained":[223],"domains,":[224],"supports":[228],"fit,":[231],"improves":[234],"benchmark":[235],"accuracy":[236],"11%":[238],"average":[240],"best":[243],"nonlinear":[244],"baseline.":[245]},"counts_by_year":[],"updated_date":"2026-11-01T11:40:01.658920","created_date":"2026-10-29T00:00:00"}