{"id":"https://openalex.org/W4403345973","doi":"https://doi.org/10.48550/arxiv.2410.07170","title":"Parameter Efficient Fine-tuning via Explained Variance Adaptation","display_name":"Parameter Efficient Fine-tuning via Explained Variance Adaptation","publication_year":2024,"publication_date":"2024-10-09","ids":{"openalex":"https://openalex.org/W4403345973","doi":"https://doi.org/10.48550/arxiv.2410.07170"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2410.07170","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2410.07170","pdf_url":"https://arxiv.org/pdf/2410.07170","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/2410.07170","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5034004608","display_name":"Fabian Paischer","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Paischer, Fabian","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5109819269","display_name":"Lukas Hauzenberger","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hauzenberger, Lukas","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5052543224","display_name":"Thomas Schmied","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Schmied, Thomas","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5047523909","display_name":"Benedikt Alkin","orcid":"https://orcid.org/0009-0001-6006-4837"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Alkin, Benedikt","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5001763022","display_name":"Marc Peter Deisenroth","orcid":"https://orcid.org/0000-0003-1503-680X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Deisenroth, Marc Peter","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5053148274","display_name":"Sepp Hochreiter","orcid":"https://orcid.org/0000-0001-7449-2528"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hochreiter, Sepp","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/T10531","display_name":"Advanced Vision and Imaging","score":0.36320000886917114,"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"}},"topics":[{"id":"https://openalex.org/T10531","display_name":"Advanced Vision and Imaging","score":0.36320000886917114,"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"}},{"id":"https://openalex.org/T10860","display_name":"Speech and Audio Processing","score":0.3183000087738037,"subfield":{"id":"https://openalex.org/subfields/1711","display_name":"Signal Processing"},"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/initialization","display_name":"Initialization","score":0.8304499387741089},{"id":"https://openalex.org/keywords/variance","display_name":"Variance (accounting)","score":0.7098438143730164},{"id":"https://openalex.org/keywords/adaptation","display_name":"Adaptation (eye)","score":0.6408167481422424},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.44273602962493896},{"id":"https://openalex.org/keywords/econometrics","display_name":"Econometrics","score":0.3960362374782562},{"id":"https://openalex.org/keywords/economics","display_name":"Economics","score":0.2622504234313965},{"id":"https://openalex.org/keywords/psychology","display_name":"Psychology","score":0.24935492873191833},{"id":"https://openalex.org/keywords/neuroscience","display_name":"Neuroscience","score":0.04908740520477295}],"concepts":[{"id":"https://openalex.org/C114466953","wikidata":"https://www.wikidata.org/wiki/Q6034165","display_name":"Initialization","level":2,"score":0.8304499387741089},{"id":"https://openalex.org/C196083921","wikidata":"https://www.wikidata.org/wiki/Q7915758","display_name":"Variance (accounting)","level":2,"score":0.7098438143730164},{"id":"https://openalex.org/C139807058","wikidata":"https://www.wikidata.org/wiki/Q352374","display_name":"Adaptation (eye)","level":2,"score":0.6408167481422424},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.44273602962493896},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.3960362374782562},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.2622504234313965},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.24935492873191833},{"id":"https://openalex.org/C169760540","wikidata":"https://www.wikidata.org/wiki/Q207011","display_name":"Neuroscience","level":1,"score":0.04908740520477295},{"id":"https://openalex.org/C121955636","wikidata":"https://www.wikidata.org/wiki/Q4116214","display_name":"Accounting","level":1,"score":0.0},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2410.07170","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2410.07170","pdf_url":"https://arxiv.org/pdf/2410.07170","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.2410.07170","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2410.07170","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:2410.07170","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2410.07170","pdf_url":"https://arxiv.org/pdf/2410.07170","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":[{"id":"https://openalex.org/G1746236357","display_name":null,"funder_award_id":"10.55776/FG9","funder_id":"https://openalex.org/F4320321181","funder_display_name":"Austrian Science Fund"},{"id":"https://openalex.org/G6276076784","display_name":null,"funder_award_id":"FG 9-N","funder_id":"https://openalex.org/F4320321181","funder_display_name":"Austrian Science Fund"}],"funders":[{"id":"https://openalex.org/F4320321181","display_name":"Austrian Science Fund","ror":"https://ror.org/013tf3c58"}],"has_content":{"grobid_xml":false,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4403345973.pdf"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2748952813","https://openalex.org/W3204184292","https://openalex.org/W3176564347","https://openalex.org/W1985458517","https://openalex.org/W2355833770","https://openalex.org/W3031039437","https://openalex.org/W183202219","https://openalex.org/W3095877357","https://openalex.org/W2072565696"],"abstract_inverted_index":{"Foundation":[0],"models":[1],"(FMs)":[2],"are":[3],"pre-trained":[4],"on":[5,37,97],"large-scale":[6],"datasets":[7],"and":[8,90,102,150,154,163,203],"then":[9],"fine-tuned":[10],"for":[11,33,60,107,122],"a":[12,123,142,170,190],"specific":[13],"downstream":[14],"task.":[15],"The":[16],"most":[17,81,120],"common":[18],"fine-tuning":[19,145],"method":[20],"is":[21,58],"to":[22,141,195],"update":[23],"pretrained":[24],"weights":[25],"via":[26],"low-rank":[27],"adaptation":[28],"(LoRA).":[29],"Existing":[30],"initialization":[31,73,108,198],"strategies":[32],"LoRA":[34,197],"often":[35],"rely":[36],"singular":[38],"value":[39],"decompositions":[40],"(SVD)":[41],"of":[42,99,135,144,172,180],"gradients":[43],"or":[44],"weight":[45],"matrices.":[46],"However,":[47],"they":[48,110],"do":[49],"not":[50],"provably":[51,84],"maximize":[52],"the":[53,77,80,86,104,115,119,133,165,178],"expected":[54,87],"gradient":[55,88],"signal,":[56],"which":[57],"critical":[59],"fast":[61],"adaptation.":[62],"To":[63],"this":[64],"end,":[65],"we":[66],"introduce":[67],"Explained":[68],"Variance":[69],"Adaptation":[70],"(EVA),":[71],"an":[72],"scheme":[74],"that":[75,117,131],"uses":[76],"directions":[78,116],"capturing":[79],"activation":[82,100],"variance,":[83],"maximizing":[85],"signal":[89],"accelerating":[91],"fine-tuning.":[92],"EVA":[93,127,140,157,188],"performs":[94],"incremental":[95],"SVD":[96],"minibatches":[98],"vectors":[101,106],"selects":[103],"right-singular":[105],"once":[109],"converged.":[111],"Further,":[112],"by":[113],"selecting":[114],"capture":[118],"activation-variance":[121],"given":[124],"rank":[125,184],"budget,":[126],"accommodates":[128],"adaptive":[129],"ranks":[130],"reduce":[132],"number":[134,179],"trainable":[136,181],"parameters.":[137],"We":[138],"apply":[139],"variety":[143],"tasks":[146,173],"as":[147],"language":[148],"generation":[149],"understanding,":[151],"image":[152],"classification,":[153],"reinforcement":[155],"learning.":[156],"exhibits":[158],"faster":[159],"convergence":[160],"than":[161],"competitors":[162],"achieves":[164],"highest":[166],"average":[167],"score":[168],"across":[169],"multitude":[171],"per":[174],"domain":[175],"while":[176],"reducing":[177],"parameters":[182],"through":[183],"redistribution.":[185],"In":[186],"summary,":[187],"establishes":[189],"new":[191],"Pareto":[192],"frontier":[193],"compared":[194],"existing":[196],"schemes":[199],"in":[200],"both":[201],"accuracy":[202],"efficiency.":[204]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2024-10-12T00:00:00"}