{"id":"https://openalex.org/W4416034175","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.1156","title":"CLIP-UP: A Simple and Efficient Mixture-of-Experts CLIP Training Recipe with Sparse Upcycling","display_name":"CLIP-UP: A Simple and Efficient Mixture-of-Experts CLIP Training Recipe with Sparse Upcycling","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416034175","doi":"https://doi.org/10.18653/v1/2025.findings-emnlp.1156"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.1156","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1156","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1156.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.1156.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5021260264","display_name":"Xinze Wang","orcid":"https://orcid.org/0000-0002-8233-4260"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xinze Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104666046","display_name":"Chen Chen","orcid":"https://orcid.org/0000-0001-7179-0861"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen Chen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5112593580","display_name":"Yinfei Yang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yinfei Yang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5012764388","display_name":"Hong-You Chen","orcid":"https://orcid.org/0000-0002-8127-5588"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hong-You Chen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100385138","display_name":"Bowen Zhang","orcid":"https://orcid.org/0000-0002-3581-9476"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bowen Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5052127251","display_name":"Aditya Kumar Pal","orcid":"https://orcid.org/0000-0003-3005-7668"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Aditya Pal","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5111419785","display_name":"Xiangxin Zhu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiangxin Zhu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5017048711","display_name":"Xianzhi Du","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xianzhi Du","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":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"21186","last_page":"21200"},"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":0.034699998795986176,"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":0.034699998795986176,"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":0.03440000116825104,"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/T10789","display_name":"Interactive and Immersive Displays","score":0.03280000016093254,"subfield":{"id":"https://openalex.org/subfields/1709","display_name":"Human-Computer Interaction"},"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/recipe","display_name":"Recipe","score":0.7910000085830688},{"id":"https://openalex.org/keywords/simple","display_name":"Simple (philosophy)","score":0.6331999897956848},{"id":"https://openalex.org/keywords/training","display_name":"Training (meteorology)","score":0.593999981880188},{"id":"https://openalex.org/keywords/training-set","display_name":"Training set","score":0.27639999985694885}],"concepts":[{"id":"https://openalex.org/C2778671685","wikidata":"https://www.wikidata.org/wiki/Q219239","display_name":"Recipe","level":2,"score":0.7910000085830688},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6496000289916992},{"id":"https://openalex.org/C2780586882","wikidata":"https://www.wikidata.org/wiki/Q7520643","display_name":"Simple (philosophy)","level":2,"score":0.6331999897956848},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.593999981880188},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.48510000109672546},{"id":"https://openalex.org/C199639397","wikidata":"https://www.wikidata.org/wiki/Q1788588","display_name":"Engineering drawing","level":1,"score":0.32100000977516174},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2809999883174896},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.27639999985694885},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.25040000677108765},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.2287999987602234}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-emnlp.1156","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1156","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1156.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"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-emnlp.1156","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-emnlp.1156","pdf_url":"https://aclanthology.org/2025.findings-emnlp.1156.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":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416034175.pdf","grobid_xml":"https://content.openalex.org/works/W4416034175.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Mixture-of-Experts":[0],"(MoE)":[1],"models":[2,17,24],"are":[3],"crucial":[4],"for":[5,130],"scaling":[6],"model":[7,44,98],"capacity":[8],"while":[9,102],"controlling":[10],"inference":[11,108],"costs.While":[12],"integrating":[13],"MoE":[14,48],"into":[15,45],"multimodal":[16],"like":[18],"CLIP":[19,43,70,96,134],"improves":[20],"performance,":[21],"training":[22,36,64,116],"these":[23],"is":[25],"notoriously":[26],"challenging":[27],"and":[28,55,66,82,86,127],"expensive.We":[29],"propose":[30],"CLIP-Upcycling":[31],"(CLIP-UP),":[32],"an":[33],"efficient":[34],"alternative":[35],"strategy":[37],"that":[38,60],"converts":[39],"a":[40,46,125],"pre-trained":[41],"dense":[42,78],"sparse":[47,69,122],"architecture.Through":[49],"extensive":[50],"experimentation":[51],"with":[52,74],"various":[53],"settings":[54],"auxiliary":[56],"losses,":[57],"we":[58],"demonstrate":[59,111],"CLIP-UP":[61],"significantly":[62],"reduces":[63],"complexity":[65],"cost.Remarkably,":[67],"our":[68,115],"B/16":[71],"model,":[72],"trained":[73],"CLIP-UP,":[75],"outperforms":[76],"its":[77],"counterpart":[79],"by":[80],"7.2%":[81],"6.6%":[83],"on":[84,99],"COCO":[85],"Flickr30k":[87],"text-toimage":[88],"Recall@1":[89],"benchmarks":[90],"respectively.It":[91],"even":[92],"surpasses":[93],"the":[94,107,112],"larger":[95],"L/14":[97],"this":[100],"task":[101],"using":[103],"only":[104],"30%":[105],"of":[106,114],"FLOPs.We":[109],"further":[110],"generalizability":[113],"recipe":[117],"across":[118],"different":[119],"scales,":[120],"establishing":[121],"upcycling":[123],"as":[124],"practical":[126],"scalable":[128],"approach":[129],"building":[131],"efficient,":[132],"high-performance":[133],"models.":[135]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-11-08T00:00:00"}