{"id":"https://openalex.org/W7102407997","doi":"https://doi.org/10.48550/arxiv.2510.24038","title":"Enhancing CLIP Robustness via Cross-Modality Alignment","display_name":"Enhancing CLIP Robustness via Cross-Modality Alignment","publication_year":2025,"publication_date":"2025-10-28","ids":{"openalex":"https://openalex.org/W7102407997","doi":"https://doi.org/10.48550/arxiv.2510.24038"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2510.24038","doi":"https://doi.org/10.48550/arxiv.2510.24038","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2510.24038","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"volume":null,"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2510.24038","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5127808960","display_name":"Xingyu Zhu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhu, Xingyu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Zhu, Beier","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhu, Beier","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Wang, Shuo","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Shuo","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5124909181","display_name":"Kesen Zhao","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Kesen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":null,"display_name":"Zhang, Hanwang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Hanwang","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":true,"primary_topic":{"id":"https://openalex.org/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9955999851226807,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.9955999851226807,"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":0.0012000000569969416,"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":0.0005000000237487257,"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/clip","display_name":"CLIP","score":0.9079999923706055},{"id":"https://openalex.org/keywords/adversarial-robustness","display_name":"adversarial robustness","score":0.9020000100135803},{"id":"https://openalex.org/keywords/cross-modal-alignment","display_name":"cross-modal alignment","score":0.7620000243186951},{"id":"https://openalex.org/keywords/optimal-transport","display_name":"optimal transport","score":0.7739999890327454},{"id":"https://openalex.org/keywords/vision-language-models","display_name":"vision-language models","score":0.8289999961853027},{"id":"https://openalex.org/keywords/zero-shot-classification","display_name":"zero-shot classification","score":0.921999990940094},{"id":"https://openalex.org/keywords/adversarial-perturbations","display_name":"adversarial perturbations","score":0.7929999828338623},{"id":"https://openalex.org/keywords/pgd-attacks","display_name":"PGD attacks","score":0.5849999785423279},{"id":"https://openalex.org/keywords/image-text-embeddings","display_name":"image-text embeddings","score":0.3160000145435333}],"concepts":[{"id":"https://openalex.org/C37736160","wikidata":"https://www.wikidata.org/wiki/Q1801315","display_name":"Adversarial system","level":2,"score":0.8695999979972839},{"id":"https://openalex.org/C32834561","wikidata":"https://www.wikidata.org/wiki/Q660730","display_name":"Subspace topology","level":2,"score":0.6815999746322632},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6812999844551086},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.6784999966621399},{"id":"https://openalex.org/C97931131","wikidata":"https://www.wikidata.org/wiki/Q5282087","display_name":"Discriminative model","level":2,"score":0.6488999724388123},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.609499990940094},{"id":"https://openalex.org/C2776436953","wikidata":"https://www.wikidata.org/wiki/Q5163215","display_name":"Consistency (knowledge bases)","level":2,"score":0.527899980545044},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.5044000148773193},{"id":"https://openalex.org/C115961682","wikidata":"https://www.wikidata.org/wiki/Q860623","display_name":"Image (mathematics)","level":2,"score":0.4586000144481659},{"id":"https://openalex.org/C192209626","wikidata":"https://www.wikidata.org/wiki/Q190909","display_name":"Focus (optics)","level":2,"score":0.44830000400543213},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.4302999973297119},{"id":"https://openalex.org/C75294576","wikidata":"https://www.wikidata.org/wiki/Q5165192","display_name":"Contextual image classification","level":3,"score":0.42100000381469727},{"id":"https://openalex.org/C57493831","wikidata":"https://www.wikidata.org/wiki/Q3134666","display_name":"Projection (relational algebra)","level":2,"score":0.3783999979496002},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.375},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.35040000081062317},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.3409999907016754},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.33079999685287476},{"id":"https://openalex.org/C52622490","wikidata":"https://www.wikidata.org/wiki/Q1026626","display_name":"Feature extraction","level":2,"score":0.3249000012874603},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.32359999418258667},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.31940001249313354},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.29829999804496765},{"id":"https://openalex.org/C83665646","wikidata":"https://www.wikidata.org/wiki/Q42139305","display_name":"Feature vector","level":2,"score":0.273499995470047},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.26980000734329224},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.266400009393692},{"id":"https://openalex.org/C12362212","wikidata":"https://www.wikidata.org/wiki/Q728435","display_name":"Linear subspace","level":2,"score":0.25760000944137573},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.2517000138759613}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2510.24038","doi":"https://doi.org/10.48550/arxiv.2510.24038","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2510.24038","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2510.24038","doi":"https://doi.org/10.48550/arxiv.2510.24038","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2510.24038","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[{"id":"https://metadata.un.org/sdg/10","display_name":"Reduced inequalities","score":0.7294805645942688}],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Vision-language":[0],"models":[1,128],"(VLMs)":[2],"such":[3],"as":[4,41,132],"CLIP":[5],"demonstrate":[6,181],"strong":[7],"generalization":[8],"in":[9,34,65,97],"zero-shot":[10,178],"classification":[11,66,179],"but":[12],"remain":[13],"highly":[14],"vulnerable":[15],"to":[16,62],"adversarial":[17,24,59,85,105,163,200],"perturbations.":[18],"Existing":[19],"methods":[20],"primarily":[21],"focus":[22],"on":[23,193,206],"fine-tuning":[25],"or":[26],"prompt":[27],"optimization;":[28],"they":[29],"often":[30],"overlook":[31],"the":[32,42,98,146,152,182],"gaps":[33],"CLIP's":[35],"encoded":[36],"features,":[37,115],"which":[38],"is":[39,55,166],"shown":[40],"text":[43,114],"and":[44,93,130,139,168,195],"image":[45,106],"features":[46],"lie":[47],"far":[48],"apart":[49],"from":[50],"each":[51],"other.":[52],"This":[53,155],"misalignment":[54,86],"significantly":[56],"amplified":[57],"under":[58,162,198],"perturbations,":[60],"leading":[61],"severe":[63],"degradation":[64],"performance.":[67],"To":[68],"address":[69],"this":[70],"problem,":[71],"we":[72],"propose":[73],"Cross-modality":[74],"Alignment,":[75],"dubbed":[76],"COLA,":[77,185],"an":[78,188],"optimal":[79],"transport-based":[80],"framework":[81],"that":[82],"explicitly":[83],"addresses":[84],"by":[87,112],"restoring":[88],"both":[89],"global":[90],"image-text":[91],"alignment":[92,142,160],"local":[94],"structural":[95],"consistency":[96],"feature":[99],"space.":[100],"(1)":[101],"COLA":[102,165],"first":[103],"projects":[104],"embeddings":[107],"onto":[108],"a":[109],"subspace":[110,147],"spanned":[111],"class":[113],"effectively":[116],"filtering":[117],"out":[118],"non-semantic":[119],"distortions":[120],"while":[121,202],"preserving":[122],"discriminative":[123],"information.":[124],"(2)":[125],"It":[126],"then":[127],"images":[129],"texts":[131],"discrete":[133],"distributions":[134],"over":[135],"multiple":[136],"augmented":[137],"views":[138],"refines":[140],"their":[141],"via":[143],"OT,":[144],"with":[145,170,187],"projection":[148],"seamlessly":[149],"integrated":[150],"into":[151],"cost":[153],"computation.":[154],"design":[156],"ensures":[157],"stable":[158],"cross-modal":[159],"even":[161],"conditions.":[164],"training-free":[167],"compatible":[169],"existing":[171],"fine-tuned":[172],"models.":[173],"Extensive":[174],"evaluations":[175],"across":[176],"14":[177],"benchmarks":[180],"effectiveness":[183],"of":[184,191],"especially":[186],"average":[189],"improvement":[190],"6.7%":[192],"ImageNet":[194],"its":[196],"variants":[197],"PGD":[199],"attacks,":[201],"maintaining":[203],"high":[204],"accuracy":[205],"clean":[207],"samples.":[208]},"counts_by_year":[],"updated_date":"2026-10-14T16:39:01.820549","created_date":"2025-10-30T00:00:00"}