{"id":"https://openalex.org/W4406880076","doi":"https://doi.org/10.48550/arxiv.2501.15065","title":"Task Arithmetic in Trust Region: A Training-Free Model Merging Approach to Navigate Knowledge Conflicts","display_name":"Task Arithmetic in Trust Region: A Training-Free Model Merging Approach to Navigate Knowledge Conflicts","publication_year":2025,"publication_date":"2025-01-25","ids":{"openalex":"https://openalex.org/W4406880076","doi":"https://doi.org/10.48550/arxiv.2501.15065"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2501.15065","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2501.15065","pdf_url":"https://arxiv.org/pdf/2501.15065","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":"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/2501.15065","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5075882217","display_name":"Wenju Sun","orcid":"https://orcid.org/0000-0002-0317-3922"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Sun, Wenju","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5005350496","display_name":"Qingyong Li","orcid":"https://orcid.org/0000-0002-3860-4809"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Qingyong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101923516","display_name":"Wen Wang","orcid":"https://orcid.org/0000-0002-5768-8016"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Wen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5018191388","display_name":"Yangli\u2010ao Geng","orcid":"https://orcid.org/0000-0003-0173-4041"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Geng, Yangli-ao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5100732746","display_name":"Boyang Li","orcid":"https://orcid.org/0000-0002-6230-2376"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Boyang","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":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/T11303","display_name":"Bayesian Modeling and Causal Inference","score":0.7807000279426575,"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/T11303","display_name":"Bayesian Modeling and Causal Inference","score":0.7807000279426575,"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/T11182","display_name":"Auction Theory and Applications","score":0.6794000267982483,"subfield":{"id":"https://openalex.org/subfields/1803","display_name":"Management Science and Operations Research"},"field":{"id":"https://openalex.org/fields/18","display_name":"Decision Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10462","display_name":"Reinforcement Learning in Robotics","score":0.670199990272522,"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/task","display_name":"Task (project management)","score":0.8310900330543518},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.6287329196929932},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.5030707716941833},{"id":"https://openalex.org/keywords/arithmetic","display_name":"Arithmetic","score":0.47835296392440796},{"id":"https://openalex.org/keywords/knowledge-management","display_name":"Knowledge management","score":0.36463767290115356},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.17771971225738525},{"id":"https://openalex.org/keywords/management","display_name":"Management","score":0.10137444734573364},{"id":"https://openalex.org/keywords/geography","display_name":"Geography","score":0.09200289845466614},{"id":"https://openalex.org/keywords/economics","display_name":"Economics","score":0.07545492053031921}],"concepts":[{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.8310900330543518},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6287329196929932},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.5030707716941833},{"id":"https://openalex.org/C94375191","wikidata":"https://www.wikidata.org/wiki/Q11205","display_name":"Arithmetic","level":1,"score":0.47835296392440796},{"id":"https://openalex.org/C56739046","wikidata":"https://www.wikidata.org/wiki/Q192060","display_name":"Knowledge management","level":1,"score":0.36463767290115356},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.17771971225738525},{"id":"https://openalex.org/C187736073","wikidata":"https://www.wikidata.org/wiki/Q2920921","display_name":"Management","level":1,"score":0.10137444734573364},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.09200289845466614},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.07545492053031921},{"id":"https://openalex.org/C153294291","wikidata":"https://www.wikidata.org/wiki/Q25261","display_name":"Meteorology","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2501.15065","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2501.15065","pdf_url":"https://arxiv.org/pdf/2501.15065","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2501.15065","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2501.15065","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":"pmh:oai:arXiv.org:2501.15065","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2501.15065","pdf_url":"https://arxiv.org/pdf/2501.15065","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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":false},"content_urls":{"pdf":"https://content.openalex.org/works/W4406880076.pdf"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W230091440","https://openalex.org/W2390279801","https://openalex.org/W2233261550","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2810751659"],"abstract_inverted_index":{"Multi-task":[0],"model":[1,42,87,112,163,236],"merging":[2,109,187,237],"offers":[3],"an":[4,203,240],"efficient":[5],"solution":[6],"for":[7,114],"integrating":[8],"knowledge":[9,98,196],"from":[10,44,127],"multiple":[11],"fine-tuned":[12,46,113],"models,":[13],"mitigating":[14],"the":[15,40,45,54,69,78,102,128,136,156,162,173,182,230],"significant":[16],"computational":[17],"and":[18,60,65,206],"storage":[19],"demands":[20],"associated":[21],"with":[22,110,135,177,211],"multi-task":[23,231],"training.":[24],"As":[25],"a":[26,111,207,212],"key":[27],"technique":[28],"in":[29,49,150,161,181],"this":[30,90,95,189],"field,":[31],"Task":[32,148],"Arithmetic":[33,149],"(TA)":[34],"defines":[35,155],"task":[36,47,58,79,107,131,174],"vectors":[37,59,132],"by":[38,101,239],"subtracting":[39],"pre-trained":[41],"($\u03b8_{\\text{pre}}$)":[43],"models":[48],"parameter":[50,164,186],"space,":[51],"then":[52],"adjusting":[53],"weight":[55],"between":[56],"these":[57,123],"$\u03b8_{\\text{pre}}$":[61],"to":[62,172],"balance":[63],"task-generalized":[64],"task-specific":[66,139,183],"knowledge.":[67],"Despite":[68],"promising":[70],"performance":[71,103,232],"of":[72,105,130,138,215,233],"TA,":[73],"conflicts":[74,124],"can":[75,193],"arise":[76],"among":[77],"vectors,":[80],"particularly":[81],"when":[82],"different":[83],"tasks":[84],"require":[85],"distinct":[86,223],"adaptations.":[88],"In":[89],"paper,":[91],"we":[92,120,146],"formally":[93],"define":[94],"issue":[96],"as":[97,159,201],"conflicts,":[99],"characterized":[100],"degradation":[104],"one":[106],"after":[108],"another":[115],"task.":[116],"Through":[117],"in-depth":[118],"analysis,":[119],"show":[121],"that":[122,133,166,227],"stem":[125],"primarily":[126],"components":[129,176],"align":[134],"gradient":[137,178],"losses":[140],"at":[141],"$\u03b8_{\\text{pre}}$.":[142],"To":[143],"address":[144],"this,":[145],"propose":[147],"Trust":[151],"Region":[152],"(TATR),":[153],"which":[154],"trust":[157,190],"region":[158],"dimensions":[160],"space":[165],"cause":[167],"only":[168],"small":[169],"changes":[170],"(corresponding":[171],"vector":[175],"orthogonal":[179],"direction)":[180],"losses.":[184],"Restricting":[185],"within":[188],"region,":[191],"TATR":[192,199,228],"effectively":[194],"alleviate":[195],"conflicts.":[197],"Moreover,":[198],"serves":[200],"both":[202],"independent":[204],"approach":[205],"plug-and-play":[208],"module":[209],"compatible":[210],"wide":[213],"range":[214],"TA-based":[216,235],"methods.":[217],"Extensive":[218],"empirical":[219],"evaluations":[220],"on":[221],"eight":[222],"datasets":[224],"robustly":[225],"demonstrate":[226],"improves":[229],"several":[234],"methods":[238],"observable":[241],"margin.":[242]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2025-10-10T00:00:00"}