{"id":"https://openalex.org/W7127098664","doi":"https://doi.org/10.48550/arxiv.2601.23155","title":"SPICE: Submodular Penalized Information-Conflict Selection for Efficient Large Language Model Training","display_name":"SPICE: Submodular Penalized Information-Conflict Selection for Efficient Large Language Model Training","publication_year":2026,"publication_date":"2026-01-30","ids":{"openalex":"https://openalex.org/W7127098664","doi":"https://doi.org/10.48550/arxiv.2601.23155"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2601.23155","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2601.23155","pdf_url":"https://arxiv.org/pdf/2601.23155","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","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2601.23155","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5124862977","display_name":"Powei Chang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chang, Powei","raw_affiliation_strings":["Bilibili Inc"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Bilibili Inc","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100685019","display_name":"Jinpeng Zhang","orcid":"https://orcid.org/0000-0003-0524-9190","observed_orcids":["https://orcid.org/0000-0003-0524-9190"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Jinpeng","raw_affiliation_strings":["Bilibili Inc"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Bilibili Inc","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124816123","display_name":"Bowen Chen","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Bowen","raw_affiliation_strings":["Bilibili Inc"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Bilibili Inc","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124864992","display_name":"Chenyu Wang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Chenyu","raw_affiliation_strings":["Bilibili Inc"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Bilibili Inc","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101298761","display_name":"Chenlu Guo","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Guo, Chenlu","raw_affiliation_strings":["Bilibili Inc"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Bilibili Inc","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5054992093","display_name":"Yixing Zhang","orcid":"https://orcid.org/0000-0002-5543-2682","observed_orcids":["https://orcid.org/0000-0002-5543-2682","https://orcid.org/0009-0009-9522-0877"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Yixing","raw_affiliation_strings":["Bilibili Inc"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Bilibili Inc","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124821208","display_name":"Yukang Gao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gao, Yukang","raw_affiliation_strings":["Bilibili Inc"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Bilibili Inc","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5041698863","display_name":"Jianxiang Xiang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiang, JianXiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124865535","display_name":"Yue Gao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gao, Yue","raw_affiliation_strings":["Bilibili Inc"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Bilibili Inc","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5075196803","display_name":"C. Sun","orcid":"https://orcid.org/0009-0004-1401-7767","observed_orcids":["https://orcid.org/0009-0004-1401-7767"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sun, Chaoqun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100627355","display_name":"Yiyi Chen","orcid":"https://orcid.org/0000-0001-9977-8960","observed_orcids":["https://orcid.org/0000-0001-9977-8960","https://orcid.org/0009-0000-2482-6446"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Yiyi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5039756970","display_name":"Dongying Kong","orcid":"https://orcid.org/0009-0002-9100-0964","observed_orcids":["https://orcid.org/0009-0002-9100-0964"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kong, Dongying","raw_affiliation_strings":["Bilibili Inc"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Bilibili Inc","institution_ids":[]}]}],"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":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":0.44359999895095825,"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":0.44359999895095825,"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":0.2410999983549118,"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":0.0722000002861023,"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/submodular-optimization","display_name":"submodular optimization","score":0.722000002861023},{"id":"https://openalex.org/keywords/fisher-information","display_name":"Fisher information","score":0.7549999952316284},{"id":"https://openalex.org/keywords/instruction-tuning","display_name":"instruction tuning","score":0.8659999966621399},{"id":"https://openalex.org/keywords/large-language-model-training","display_name":"large language model training","score":0.8420000076293945},{"id":"https://openalex.org/keywords/gradient-conflict","display_name":"gradient conflict","score":0.722000002861023},{"id":"https://openalex.org/keywords/approximation-guarantees","display_name":"approximation guarantees","score":0.3659999966621399},{"id":"https://openalex.org/keywords/early-stopping","display_name":"early stopping","score":0.7409999966621399},{"id":"https://openalex.org/keywords/proxy-models","display_name":"proxy models","score":0.8159999847412109},{"id":"https://openalex.org/keywords/training-efficiency","display_name":"training efficiency","score":0.421999990940094}],"concepts":[{"id":"https://openalex.org/C178621042","wikidata":"https://www.wikidata.org/wiki/Q7631710","display_name":"Submodular set function","level":2,"score":0.8385999798774719},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6276999711990356},{"id":"https://openalex.org/C87117476","wikidata":"https://www.wikidata.org/wiki/Q362383","display_name":"Cardinality (data modeling)","level":2,"score":0.5993000268936157},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.5436000227928162},{"id":"https://openalex.org/C2834757","wikidata":"https://www.wikidata.org/wiki/Q4925424","display_name":"Monotone polygon","level":2,"score":0.5307000279426575},{"id":"https://openalex.org/C51823790","wikidata":"https://www.wikidata.org/wiki/Q504353","display_name":"Greedy algorithm","level":2,"score":0.4887999892234802},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.4832000136375427},{"id":"https://openalex.org/C14036430","wikidata":"https://www.wikidata.org/wiki/Q3736076","display_name":"Function (biology)","level":2,"score":0.40610000491142273},{"id":"https://openalex.org/C148483581","wikidata":"https://www.wikidata.org/wiki/Q446488","display_name":"Feature selection","level":2,"score":0.38839998841285706},{"id":"https://openalex.org/C15312841","wikidata":"https://www.wikidata.org/wiki/Q1657198","display_name":"Set function","level":3,"score":0.3792000114917755},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.37860000133514404},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.3686000108718872},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.36559998989105225},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.36410000920295715},{"id":"https://openalex.org/C93959086","wikidata":"https://www.wikidata.org/wiki/Q6888345","display_name":"Model selection","level":2,"score":0.35910001397132874},{"id":"https://openalex.org/C2780148112","wikidata":"https://www.wikidata.org/wiki/Q1432581","display_name":"Proxy (statistics)","level":2,"score":0.35690000653266907},{"id":"https://openalex.org/C2776639384","wikidata":"https://www.wikidata.org/wiki/Q840396","display_name":"Ideal (ethics)","level":2,"score":0.3531999886035919},{"id":"https://openalex.org/C5465570","wikidata":"https://www.wikidata.org/wiki/Q5326898","display_name":"Early stopping","level":3,"score":0.3479999899864197},{"id":"https://openalex.org/C148764684","wikidata":"https://www.wikidata.org/wiki/Q621751","display_name":"Approximation algorithm","level":2,"score":0.34439998865127563},{"id":"https://openalex.org/C2776502983","wikidata":"https://www.wikidata.org/wiki/Q690182","display_name":"Contrast (vision)","level":2,"score":0.3140000104904175},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3095000088214874},{"id":"https://openalex.org/C2780077345","wikidata":"https://www.wikidata.org/wiki/Q16891888","display_name":"Spice","level":2,"score":0.3068000078201294},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3028999865055084},{"id":"https://openalex.org/C2776330181","wikidata":"https://www.wikidata.org/wiki/Q18358244","display_name":"Maximization","level":2,"score":0.29739999771118164},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2903999984264374},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.2825999855995178},{"id":"https://openalex.org/C2780898871","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Performance metric","level":2,"score":0.26750001311302185},{"id":"https://openalex.org/C173801870","wikidata":"https://www.wikidata.org/wiki/Q201413","display_name":"Heuristic","level":2,"score":0.267300009727478},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.2662000060081482},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.2565999925136566},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.25589999556541443}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2601.23155","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2601.23155","pdf_url":"https://arxiv.org/pdf/2601.23155","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.2601.23155","doi":"https://doi.org/10.48550/arxiv.2601.23155","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2601.23155","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":"public-domain","license_id":"https://openalex.org/licenses/public-domain","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2601.23155","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2601.23155","pdf_url":"https://arxiv.org/pdf/2601.23155","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":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7127098664.pdf","grobid_xml":"https://content.openalex.org/works/W7127098664.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4416078514","https://openalex.org/W4417299485","https://openalex.org/W4414895810","https://openalex.org/W7125102852","https://openalex.org/W4401042981","https://openalex.org/W7162441280","https://openalex.org/W6979240938","https://openalex.org/W4416035142","https://openalex.org/W4391670794","https://openalex.org/W4416037146"],"abstract_inverted_index":{"Information-based":[0],"data":[1],"selection":[2],"for":[3,116],"instruction":[4],"tuning":[5],"is":[6,44],"compelling:":[7],"maximizing":[8],"the":[9,12,51,72,149],"log-determinant":[10,55,124],"of":[11,53,62,80,148],"Fisher":[13],"information":[14,56,104,125],"yields":[15],"a":[16,25,29,45,78,99],"monotone":[17],"submodular":[18],"objective,":[19],"enabling":[20],"greedy":[21],"algorithms":[22],"to":[23],"achieve":[24],"$(1-1/e)$":[26],"approximation":[27,85],"under":[28],"cardinality":[30],"budget.":[31],"In":[32],"practice,":[33],"however,":[34],"we":[35,96],"identify":[36],"alleviating":[37],"gradient":[38],"conflicts,":[39],"misalignment":[40],"between":[41],"per-sample":[42],"gradients,":[43],"key":[46],"factor":[47],"that":[48,70,87,102,109],"slows":[49],"down":[50],"decay":[52],"marginal":[54],"gains,":[57],"thereby":[58],"preventing":[59],"significant":[60],"loss":[61],"information.":[63],"We":[64],"formalize":[65],"this":[66,94],"via":[67],"an":[68],"$\\varepsilon$-decomposition":[69],"quantifies":[71],"deviation":[73],"from":[74],"ideal":[75],"submodularity":[76],"as":[77,89],"function":[79],"conflict":[81],"statistics,":[82],"yielding":[83],"data-dependent":[84],"factors":[86],"tighten":[88],"conflicts":[90],"diminish.":[91],"Guided":[92],"by":[93],"analysis,":[95],"propose":[97],"SPICE,":[98],"conflict-aware":[100],"selector":[101],"maximizes":[103],"while":[105],"penalizing":[106],"misalignment,":[107],"and":[108,113,129,142],"supports":[110],"early":[111],"stopping":[112],"proxy":[114],"models":[115],"efficiency.":[117],"Empirically,":[118],"SPICE":[119,144],"selects":[120],"subsets":[121],"with":[122,140,164],"higher":[123],"than":[126],"original":[127],"criteria,":[128],"these":[130],"informational":[131],"gains":[132],"translate":[133],"into":[134],"performance":[135,162],"improvements:":[136],"across":[137],"8":[138],"benchmarks":[139],"LLaMA2-7B":[141],"Qwen2-7B,":[143],"uses":[145],"only":[146],"10%":[147],"data,":[150],"yet":[151],"matches":[152],"or":[153],"exceeds":[154],"6":[155],"methods":[156],"including":[157],"full-data":[158],"tuning.":[159],"This":[160],"achieves":[161],"improvements":[163],"substantially":[165],"lower":[166],"training":[167],"cost.":[168]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-02-03T00:00:00"}