{"id":"https://openalex.org/W7162759318","doi":"https://doi.org/10.48550/arxiv.2605.29380","title":"TRACER: Persistent Regularization for Robust Multimodal Finetuning","display_name":"TRACER: Persistent Regularization for Robust Multimodal Finetuning","publication_year":2026,"publication_date":"2026-05-28","ids":{"openalex":"https://openalex.org/W7162759318","doi":"https://doi.org/10.48550/arxiv.2605.29380"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.29380","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.29380","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":"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.29380","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5093470865","display_name":"Hesam Asadollahzadeh","orcid":"https://orcid.org/0009-0003-6330-2401"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Asadollahzadeh, Hesam","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137315869","display_name":"Feng Liu","orcid":"https://orcid.org/0009-0009-7351-7266"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Liu, Feng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133566538","display_name":"Christopher Leckie","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Leckie, Christopher","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135355733","display_name":"Sarah M. Erfani","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Erfani, Sarah M.","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":false,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.7116000056266785,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.7116000056266785,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.10980000346899033,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.055799998342990875,"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"}}],"keywords":[{"id":"https://openalex.org/keywords/regularization","display_name":"Regularization (linguistics)","score":0.5963000059127808},{"id":"https://openalex.org/keywords/hyperparameter","display_name":"Hyperparameter","score":0.5375000238418579},{"id":"https://openalex.org/keywords/subspace-topology","display_name":"Subspace topology","score":0.520799994468689},{"id":"https://openalex.org/keywords/computation","display_name":"Computation","score":0.392300009727478},{"id":"https://openalex.org/keywords/convergence","display_name":"Convergence (economics)","score":0.3847000002861023},{"id":"https://openalex.org/keywords/heuristics","display_name":"Heuristics","score":0.3693000078201294},{"id":"https://openalex.org/keywords/matrix-norm","display_name":"Matrix norm","score":0.35920000076293945}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6991000175476074},{"id":"https://openalex.org/C2776135515","wikidata":"https://www.wikidata.org/wiki/Q17143721","display_name":"Regularization (linguistics)","level":2,"score":0.5963000059127808},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5546000003814697},{"id":"https://openalex.org/C8642999","wikidata":"https://www.wikidata.org/wiki/Q4171168","display_name":"Hyperparameter","level":2,"score":0.5375000238418579},{"id":"https://openalex.org/C32834561","wikidata":"https://www.wikidata.org/wiki/Q660730","display_name":"Subspace topology","level":2,"score":0.520799994468689},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4262000024318695},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4113999903202057},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.392300009727478},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.3847000002861023},{"id":"https://openalex.org/C127705205","wikidata":"https://www.wikidata.org/wiki/Q5748245","display_name":"Heuristics","level":2,"score":0.3693000078201294},{"id":"https://openalex.org/C92207270","wikidata":"https://www.wikidata.org/wiki/Q939253","display_name":"Matrix norm","level":3,"score":0.35920000076293945},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.3587000072002411},{"id":"https://openalex.org/C42355184","wikidata":"https://www.wikidata.org/wiki/Q1361088","display_name":"Matrix decomposition","level":3,"score":0.3513000011444092},{"id":"https://openalex.org/C174576160","wikidata":"https://www.wikidata.org/wiki/Q1183700","display_name":"Deconvolution","level":2,"score":0.33079999685287476},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.31049999594688416},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.3068999946117401},{"id":"https://openalex.org/C187834632","wikidata":"https://www.wikidata.org/wiki/Q188804","display_name":"Factorization","level":2,"score":0.30309998989105225},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.2994000017642975},{"id":"https://openalex.org/C117251300","wikidata":"https://www.wikidata.org/wiki/Q1849855","display_name":"Parametric statistics","level":2,"score":0.2833000123500824},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.26429998874664307},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.26190000772476196}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.29380","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.29380","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.29380","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.29380","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":"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"},"sustainable_development_goals":[{"display_name":"Quality Education","score":0.623918354511261,"id":"https://metadata.un.org/sdg/4"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Mainstream":[0],"strategies":[1],"for":[2,26,37,119],"finetuning":[3,135],"pretrained":[4,58],"multimodal":[5,27],"models":[6],"often":[7],"degrade":[8],"out-of-distribution":[9],"(OOD)":[10],"robustness,":[11],"a":[12,23,34,63,87,94],"phenomenon":[13],"known":[14],"as":[15],"catastrophic":[16],"forgetting.":[17],"In":[18],"this":[19],"paper,":[20],"we":[21,84],"develop":[22],"theoretical":[24],"framework":[25,41],"contrastive":[28,125],"finetuning,":[29,77],"yielding":[30],"closed-form":[31],"solutions":[32],"and":[33,101,140,147,156],"geometric":[35],"decomposition":[36],"each":[38],"strategy.":[39],"This":[40],"shows":[42],"that":[43,86,151],"self-distillation":[44],"is":[45,153,162],"more":[46],"effective":[47],"than":[48],"other":[49],"regularization":[50],"approaches":[51],"to":[52,158],"retain":[53],"the":[54,57,106],"knowledge":[55],"of":[56],"model.":[59],"Our":[60],"analysis":[61],"reveals":[62],"largely":[64],"overlooked":[65],"limitation:":[66],"standard":[67],"Exponential":[68],"Moving":[69,89],"Average":[70,90],"(EMA)":[71],"teachers,":[72],"widely":[73],"used":[74],"in":[75,105],"robust":[76,157],"suffer":[78],"from":[79],"collapse.":[80],"To":[81],"solve":[82],"this,":[83],"prove":[85],"Weighted":[88],"(WMA)":[91],"teacher":[92],"maintains":[93],"persistent":[95],"regularizing":[96],"force":[97],"over":[98],"finite":[99],"horizons":[100],"yields":[102],"bias-free":[103],"convergence":[104],"task":[107],"subspace":[108],"while":[109],"preserving":[110],"orthogonal":[111],"knowledge.":[112],"These":[113],"insights":[114],"motivate":[115],"**TRACER**":[116],"(**T**rajectory-**R**obust":[117],"**A**nchoring":[118],"**C**ontrastive":[120],"**E**ncoder":[121],"**R**egularization),":[122],"which":[123],"combines":[124],"learning":[126],"with":[127],"WMA-guided":[128],"multi-perspective":[129],"distillation.":[130],"Extensive":[131],"experiments":[132],"on":[133],"CLIP":[134],"demonstrate":[136],"consistent":[137],"OOD":[138],"accuracy":[139],"calibration":[141],"gains":[142],"across":[143],"three":[144],"backbone":[145],"architectures,":[146],"comprehensive":[148],"ablations":[149],"confirm":[150],"TRACER":[152],"both":[154],"principled":[155],"hyperparameter":[159],"choices.":[160],"Code":[161],"available":[163],"at":[164],"[https://github.com/HesamAsad/TRACER](https://github.com/HesamAsad/TRACER).":[165]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-30T00:00:00"}