{"id":"https://openalex.org/W4416034128","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.1235","title":"Lost in Embeddings: Information Loss in Vision\u2013Language Models","display_name":"Lost in Embeddings: Information Loss in Vision\u2013Language Models","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416034128","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.1235"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.1235","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1235","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1235.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-emnlp.1235.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100749175","display_name":"Wenyan Li","orcid":"https://orcid.org/0000-0003-1861-6383"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wenyan Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101066425","display_name":"Raphael Tang","orcid":"https://orcid.org/0009-0007-2873-892X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Raphael Tang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5070178016","display_name":"Chengzu Li","orcid":"https://orcid.org/0000-0002-1879-7391"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chengzu Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5064109244","display_name":"Caiqi Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Caiqi Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5014866912","display_name":"Ivan Vuli\u0107","orcid":"https://orcid.org/0000-0002-5161-5422"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ivan Vuli\u0107","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5018138946","display_name":"Anders S\u00f8gaard","orcid":"https://orcid.org/0000-0001-5250-4276"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Anders S\u00f8gaard","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":"22676","last_page":"22693"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.973800003528595,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.973800003528595,"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":0.0044999998062849045,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.002199999988079071,"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/embedding","display_name":"Embedding","score":0.7418000102043152},{"id":"https://openalex.org/keywords/representation","display_name":"Representation (politics)","score":0.6442000269889832},{"id":"https://openalex.org/keywords/projection","display_name":"Projection (relational algebra)","score":0.6435999870300293},{"id":"https://openalex.org/keywords/data-loss","display_name":"Data loss","score":0.6431000232696533},{"id":"https://openalex.org/keywords/information-loss","display_name":"Information loss","score":0.6312999725341797},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.49470001459121704},{"id":"https://openalex.org/keywords/image","display_name":"Image (mathematics)","score":0.48669999837875366},{"id":"https://openalex.org/keywords/modality","display_name":"Modality (human\u2013computer interaction)","score":0.4442000091075897},{"id":"https://openalex.org/keywords/measure","display_name":"Measure (data warehouse)","score":0.4156000018119812}],"concepts":[{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.7418000102043152},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6547999978065491},{"id":"https://openalex.org/C2776359362","wikidata":"https://www.wikidata.org/wiki/Q2145286","display_name":"Representation (politics)","level":3,"score":0.6442000269889832},{"id":"https://openalex.org/C57493831","wikidata":"https://www.wikidata.org/wiki/Q3134666","display_name":"Projection (relational algebra)","level":2,"score":0.6435999870300293},{"id":"https://openalex.org/C193519340","wikidata":"https://www.wikidata.org/wiki/Q891179","display_name":"Data loss","level":2,"score":0.6431000232696533},{"id":"https://openalex.org/C2988416141","wikidata":"https://www.wikidata.org/wiki/Q6031139","display_name":"Information loss","level":2,"score":0.6312999725341797},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6184999942779541},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.49470001459121704},{"id":"https://openalex.org/C115961682","wikidata":"https://www.wikidata.org/wiki/Q860623","display_name":"Image (mathematics)","level":2,"score":0.48669999837875366},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.48570001125335693},{"id":"https://openalex.org/C2780226545","wikidata":"https://www.wikidata.org/wiki/Q6888030","display_name":"Modality (human\u2013computer interaction)","level":2,"score":0.4442000091075897},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.4156000018119812},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.36059999465942383},{"id":"https://openalex.org/C2776502983","wikidata":"https://www.wikidata.org/wiki/Q690182","display_name":"Contrast (vision)","level":2,"score":0.35040000081062317},{"id":"https://openalex.org/C2778572836","wikidata":"https://www.wikidata.org/wiki/Q380933","display_name":"Space (punctuation)","level":2,"score":0.34850001335144043},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.33980000019073486},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.32919999957084656},{"id":"https://openalex.org/C207363949","wikidata":"https://www.wikidata.org/wiki/Q462915","display_name":"Visual space","level":3,"score":0.31859999895095825},{"id":"https://openalex.org/C36464697","wikidata":"https://www.wikidata.org/wiki/Q451553","display_name":"Visualization","level":2,"score":0.3075000047683716},{"id":"https://openalex.org/C141379421","wikidata":"https://www.wikidata.org/wiki/Q6094427","display_name":"Iterative reconstruction","level":2,"score":0.27079999446868896},{"id":"https://openalex.org/C67186912","wikidata":"https://www.wikidata.org/wiki/Q367664","display_name":"Data modeling","level":2,"score":0.2705000042915344},{"id":"https://openalex.org/C61797465","wikidata":"https://www.wikidata.org/wiki/Q1188986","display_name":"Term (time)","level":2,"score":0.2662000060081482},{"id":"https://openalex.org/C160920958","wikidata":"https://www.wikidata.org/wiki/Q7662746","display_name":"Synthetic data","level":2,"score":0.26159998774528503},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.25690001249313354},{"id":"https://openalex.org/C161666082","wikidata":"https://www.wikidata.org/wiki/Q14551923","display_name":"Oblique projection","level":3,"score":0.25440001487731934},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.25290000438690186},{"id":"https://openalex.org/C2781238097","wikidata":"https://www.wikidata.org/wiki/Q175026","display_name":"Object (grammar)","level":2,"score":0.251800000667572}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.18653/v1/2025.findings-emnlp.1235","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1235","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1235.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},{"id":"pmh:oai:pure.atira.dk:openaire_cris_publications/4e449117-815c-4e6e-86de-b2dab3420be7","is_oa":true,"landing_page_url":"https://researchprofiles.ku.dk/da/publications/4e449117-815c-4e6e-86de-b2dab3420be7","pdf_url":null,"source":{"id":"https://openalex.org/S4306401983","display_name":"Research at the University of Copenhagen (University of Copenhagen)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I124055696","host_organization_name":"University of Copenhagen","host_organization_lineage":["https://openalex.org/I124055696"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Li , W , Tang , R , Li , C , Zhang , C , Vuli\u0107 , I & S\u00f8gaard , A 2025 , Lost in Embeddings : Information Loss in Vision\u2013Language Models . in Findings of the Association for Computational Linguistics: EMNLP 2025 . Association for Computational Linguistics (ACL) , pp. 22676-22693 , 2025 Conference on Empirical Methods in Natural Language Processing , Suzhou , China , 04/09/2025 . https://doi.org/10.18653/v1/2025.findings-emnlp.1235","raw_type":"contributionToPeriodical"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.1235","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1235","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1235.pdf","source":null,"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":false},"content_urls":{"pdf":"https://content.openalex.org/works/W4416034128.pdf"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Vision\u2013language":[0],"models":[1,158],"(VLMs)":[2],"often":[3],"process":[4],"visual":[5,94,117],"inputs":[6],"through":[7],"a":[8,14,23],"pretrained":[9],"vision":[10],"encoder,":[11],"followed":[12],"by":[13,36,60,72,92,123],"projection":[15,38],"into":[16],"the":[17,31,62,97,113],"language":[18],"model\u2019s":[19],"embedding":[20,134],"space":[21],"via":[22],"connector":[24],"component.":[25],"While":[26],"crucial":[27],"for":[28,139],"modality":[29],"fusion,":[30],"potential":[32],"information":[33,70,90,152],"loss":[34,59,91,101,153],"induced":[35],"this":[37,58],"step":[39],"and":[40,56,83],"its":[41],"direct":[42],"impact":[43],"on":[44,142],"model":[45,140],"capabilities":[46],"remain":[47],"understudied.":[48],"We":[49],"introduce":[50],"two":[51],"complementary":[52],"approaches":[53],"to":[54],"examine":[55],"quantify":[57],"analyzing":[61,73],"latent":[63],"representation":[64],"space.":[65],"First,":[66],"we":[67,87],"evaluate":[68],"semantic":[69],"preservation":[71],"changes":[74],"in":[75,129],"k-nearest":[76,120],"neighbor":[77],"relationships":[78],"between":[79],"image":[80,104],"representations,":[81,118],"before":[82],"after":[84],"projection.":[85],"Second,":[86],"directly":[88],"measure":[89],"reconstructing":[93],"embeddings":[95],"from":[96],"projected":[98],"representation,":[99],"localizing":[100],"at":[102],"an":[103],"patch":[105],"level.":[106],"Experiments":[107],"reveal":[108],"that":[109,148],"connectors":[110],"substantially":[111],"distort":[112],"local":[114],"geometry":[115],"of":[116,150],"with":[119,127],"neighbors":[121],"diverging":[122],"40\u201360%":[124],"post-projection,":[125],"correlating":[126],"degradation":[128],"retrieval":[130],"performance.":[131],"The":[132],"patch-level":[133],"reconstruction":[135],"provides":[136],"interpretable":[137],"insights":[138],"behavior":[141],"visually":[143],"grounded":[144],"question-answering":[145],"tasks,":[146],"finding":[147],"areas":[149],"high":[151],"reliably":[154],"predict":[155],"instances":[156],"where":[157],"struggle.":[159]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-11-08T00:00:00"}