{"id":"https://openalex.org/W7212017136","doi":"https://doi.org/10.48550/arxiv.2609.08354","title":"Geometry-Aware Bayesian Parameter-Efficient Fine-Tuning on the Stiefel Manifold via Stein Variational Gradient Descent","display_name":"Geometry-Aware Bayesian Parameter-Efficient Fine-Tuning on the Stiefel Manifold via Stein Variational Gradient Descent","publication_year":2026,"publication_date":"2026-09-08","ids":{"openalex":"https://openalex.org/W7212017136","doi":"https://doi.org/10.48550/arxiv.2609.08354"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.08354","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.08354","pdf_url":"https://arxiv.org/pdf/2609.08354","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":null,"license_id":null,"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/2609.08354","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5103228838","display_name":"Quang-Duy Tran","orcid":"https://orcid.org/0000-0002-2244-9154","observed_orcids":["https://orcid.org/0000-0002-2244-9154"]},"institutions":[{"id":"https://openalex.org/I149704539","display_name":"Deakin University","ror":"https://ror.org/02czsnj07","country_code":"AU","type":"education","lineage":["https://openalex.org/I149704539"]},{"id":"https://openalex.org/I9927081","display_name":"Aalto University","ror":"https://ror.org/020hwjq30","country_code":"FI","type":"education","lineage":["https://openalex.org/I9927081"]}],"countries":["AU","FI"],"is_corresponding":false,"raw_author_name":"Tran, Quang-Duy","raw_affiliation_strings":["Deakin Applied Artificial Intelligence Initiative , Deakin University , Geelong , Australia","Department of Computer Science , Aalto University , Espoo , Finland"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Deakin Applied Artificial Intelligence Initiative , Deakin University , Geelong , Australia","institution_ids":["https://openalex.org/I149704539"]},{"raw_affiliation_string":"Department of Computer Science , Aalto University , Espoo , Finland","institution_ids":["https://openalex.org/I9927081"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5150986616","display_name":"Trung Le","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I56590836","display_name":"Monash University","ror":"https://ror.org/02bfwt286","country_code":"AU","type":"education","lineage":["https://openalex.org/I56590836"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Le, Trung","raw_affiliation_strings":["Department of Data Science & AI , Monash University , Melbourne , Australia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Data Science & AI , Monash University , Melbourne , Australia","institution_ids":["https://openalex.org/I56590836"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5107570186","display_name":"Bao Duong","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I149704539","display_name":"Deakin University","ror":"https://ror.org/02czsnj07","country_code":"AU","type":"education","lineage":["https://openalex.org/I149704539"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Duong, Bao","raw_affiliation_strings":["Deakin Applied Artificial Intelligence Initiative , Deakin University , Geelong , Australia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Deakin Applied Artificial Intelligence Initiative , Deakin University , Geelong , Australia","institution_ids":["https://openalex.org/I149704539"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124791007","display_name":"Phuoc Nguyen","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I149704539","display_name":"Deakin University","ror":"https://ror.org/02czsnj07","country_code":"AU","type":"education","lineage":["https://openalex.org/I149704539"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Nguyen, Phuoc","raw_affiliation_strings":["Deakin Applied Artificial Intelligence Initiative , Deakin University , Geelong , Australia","Deakin Applied Artificial Intelligence Initiative , Deakin University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Deakin Applied Artificial Intelligence Initiative , Deakin University , Geelong , Australia","institution_ids":["https://openalex.org/I149704539"]},{"raw_affiliation_string":"Deakin Applied Artificial Intelligence Initiative , Deakin University","institution_ids":["https://openalex.org/I149704539"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5100705489","display_name":"Thin Nguyen","orcid":"https://orcid.org/0000-0003-3467-8963","observed_orcids":["https://orcid.org/0000-0003-3467-8963","https://orcid.org/0000-0002-9045-8589"]},"institutions":[{"id":"https://openalex.org/I149704539","display_name":"Deakin University","ror":"https://ror.org/02czsnj07","country_code":"AU","type":"education","lineage":["https://openalex.org/I149704539"]}],"countries":["AU"],"is_corresponding":false,"raw_author_name":"Nguyen, Thin","raw_affiliation_strings":["Deakin Applied Artificial Intelligence Initiative , Deakin University , Geelong , Australia"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Deakin Applied Artificial Intelligence Initiative , Deakin University , Geelong , Australia","institution_ids":["https://openalex.org/I149704539"]}]}],"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.6044999957084656,"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":0.6044999957084656,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.1289999932050705,"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":0.054499998688697815,"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/parameter-efficient-fine-tuning","display_name":"parameter-efficient fine-tuning","score":0.9229999780654907},{"id":"https://openalex.org/keywords/low-rank-adaptation","display_name":"low-rank adaptation","score":0.9179999828338623},{"id":"https://openalex.org/keywords/stiefel-manifold","display_name":"Stiefel manifold","score":0.972000002861023},{"id":"https://openalex.org/keywords/stein-variational-gradient-descent","display_name":"Stein variational gradient descent","score":0.9990000128746033},{"id":"https://openalex.org/keywords/uncertainty-quantification","display_name":"uncertainty quantification","score":0.7250000238418579},{"id":"https://openalex.org/keywords/model-calibration","display_name":"model calibration","score":0.8429999947547913},{"id":"https://openalex.org/keywords/orthogonality-constraints","display_name":"orthogonality constraints","score":0.796999990940094},{"id":"https://openalex.org/keywords/singular-value-decomposition","display_name":"singular value decomposition","score":0.6190000176429749},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.578000009059906},{"id":"https://openalex.org/keywords/bayesian-neural-networks","display_name":"Bayesian neural networks","score":0.39399999380111694}],"concepts":[{"id":"https://openalex.org/C612670","wikidata":"https://www.wikidata.org/wiki/Q7616373","display_name":"Stiefel manifold","level":2,"score":0.9660000205039978},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.6146000027656555},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.6015999913215637},{"id":"https://openalex.org/C529865628","wikidata":"https://www.wikidata.org/wiki/Q1790740","display_name":"Manifold (fluid mechanics)","level":2,"score":0.582099974155426},{"id":"https://openalex.org/C32834561","wikidata":"https://www.wikidata.org/wiki/Q660730","display_name":"Subspace topology","level":2,"score":0.5472999811172485},{"id":"https://openalex.org/C22789450","wikidata":"https://www.wikidata.org/wiki/Q420904","display_name":"Singular value decomposition","level":2,"score":0.5248000025749207},{"id":"https://openalex.org/C17137986","wikidata":"https://www.wikidata.org/wiki/Q215067","display_name":"Orthogonality","level":2,"score":0.5224000215530396},{"id":"https://openalex.org/C28826006","wikidata":"https://www.wikidata.org/wiki/Q33521","display_name":"Applied mathematics","level":1,"score":0.4814000129699707},{"id":"https://openalex.org/C129782007","wikidata":"https://www.wikidata.org/wiki/Q162886","display_name":"Euclidean geometry","level":2,"score":0.47870001196861267},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.4611000120639801},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.4316999912261963},{"id":"https://openalex.org/C186450821","wikidata":"https://www.wikidata.org/wiki/Q17295","display_name":"Euclidean space","level":2,"score":0.40610000491142273},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.39160001277923584},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.3831000030040741},{"id":"https://openalex.org/C124681953","wikidata":"https://www.wikidata.org/wiki/Q339062","display_name":"Decomposition","level":2,"score":0.3822999894618988},{"id":"https://openalex.org/C42355184","wikidata":"https://www.wikidata.org/wiki/Q1361088","display_name":"Matrix decomposition","level":3,"score":0.35199999809265137},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.34060001373291016},{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.3321000039577484},{"id":"https://openalex.org/C107673813","wikidata":"https://www.wikidata.org/wiki/Q812534","display_name":"Bayesian probability","level":2,"score":0.31619998812675476},{"id":"https://openalex.org/C13662910","wikidata":"https://www.wikidata.org/wiki/Q193139","display_name":"Trajectory","level":2,"score":0.31200000643730164},{"id":"https://openalex.org/C109282560","wikidata":"https://www.wikidata.org/wiki/Q4166054","display_name":"Singular value","level":3,"score":0.3001999855041504},{"id":"https://openalex.org/C2776637919","wikidata":"https://www.wikidata.org/wiki/Q624380","display_name":"Descent (aeronautics)","level":2,"score":0.3001999855041504},{"id":"https://openalex.org/C195065555","wikidata":"https://www.wikidata.org/wiki/Q214881","display_name":"Curvature","level":2,"score":0.2750999927520752},{"id":"https://openalex.org/C157553263","wikidata":"https://www.wikidata.org/wiki/Q5168004","display_name":"Coordinate descent","level":2,"score":0.27489998936653137},{"id":"https://openalex.org/C169391604","wikidata":"https://www.wikidata.org/wiki/Q7604402","display_name":"Statistical manifold","level":5,"score":0.27070000767707825},{"id":"https://openalex.org/C2778671503","wikidata":"https://www.wikidata.org/wiki/Q7542110","display_name":"Slow manifold","level":3,"score":0.26899999380111694},{"id":"https://openalex.org/C20483540","wikidata":"https://www.wikidata.org/wiki/Q4200516","display_name":"Invariant manifold","level":2,"score":0.26739999651908875},{"id":"https://openalex.org/C120174047","wikidata":"https://www.wikidata.org/wiki/Q847073","display_name":"Euclidean distance","level":2,"score":0.258899986743927},{"id":"https://openalex.org/C151876577","wikidata":"https://www.wikidata.org/wiki/Q7049464","display_name":"Nonlinear dimensionality reduction","level":3,"score":0.25279998779296875}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.08354","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.08354","pdf_url":"https://arxiv.org/pdf/2609.08354","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.08354","doi":"https://doi.org/10.48550/arxiv.2609.08354","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.08354","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:2609.08354","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.08354","pdf_url":"https://arxiv.org/pdf/2609.08354","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7212017136.pdf","grobid_xml":"https://content.openalex.org/works/W7212017136.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4414819810","https://openalex.org/W7131064314","https://openalex.org/W7171651333","https://openalex.org/W2963857221","https://openalex.org/W7213930786","https://openalex.org/W7077973525","https://openalex.org/W7166815373","https://openalex.org/W7213581174","https://openalex.org/W2799120050","https://openalex.org/W2934434822"],"abstract_inverted_index":{"Several":[0],"geometry-aware":[1,116],"approaches":[2],"to":[3,18,80,103],"low-rank":[4,27,95],"adaptation":[5,62],"have":[6,52],"emerged":[7],"for":[8,29],"parameter-efficient":[9],"fine-tuning":[10],"of":[11,22,26,49,78],"large":[12],"pre-trained":[13],"models.":[14],"These":[15],"methods":[16,63,157],"aim":[17],"take":[19],"full":[20],"advantage":[21],"the":[23,31,73,94,100,105,133],"geometric":[24,112],"structure":[25],"manifolds":[28],"improving":[30],"efficiency":[32],"in":[33,161],"subspace":[34],"utilization":[35],"and":[36,128,146,153],"reducing":[37],"redundancy":[38],"by":[39],"enforcing":[40],"orthogonality":[41],"constraints":[42],"during":[43,122],"optimization.":[44],"The":[45],"strong":[46,143],"empirical":[47],"results":[48],"these":[50],"techniques":[51],"motivated":[53],"further":[54],"study":[55],"into":[56],"whether":[57],"predictions":[58],"from":[59],"such":[60],"geometry-based":[61],"could":[64],"be":[65],"overconfident.":[66],"In":[67,91],"this":[68,92,115],"paper,":[69],"we":[70],"build":[71],"on":[72,85,132],"singular":[74],"value":[75],"decomposition":[76],"factorization":[77],"adapters":[79,131],"develop":[81],"a":[82],"framework":[83],"based":[84],"Stein":[86],"variational":[87],"gradient":[88],"descent":[89],"(SVGD).":[90],"formulation,":[93],"matrices":[96],"are":[97,159],"transported":[98],"along":[99],"Stiefel":[101,134],"manifold":[102],"match":[104],"targeted":[106],"distributions":[107],"while":[108],"retaining":[109],"their":[110],"crucial":[111],"structure.":[113],"Since":[114],"SVGD":[117,152],"approach":[118],"provides":[119],"multiple":[120],"solutions":[121],"inference,":[123],"it":[124],"supports":[125],"uncertainty":[126,155],"quantification":[127],"produces":[129],"better-calibrated":[130],"manifold.":[135],"Extensive":[136],"experiments":[137],"show":[138],"that":[139,158],"our":[140],"method":[141],"delivers":[142],"model":[144],"calibration":[145],"attains":[147],"higher":[148],"prediction":[149],"accuracy":[150],"than":[151],"related":[154],"estimation":[156],"formulated":[160],"Euclidean":[162],"space.":[163]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-09-10T00:00:00"}