{"id":"https://openalex.org/W7224895694","doi":"https://doi.org/10.48550/arxiv.2610.39467","title":"Calibrating Attention-Based Token Pruning for Efficient Multi-Image Synthesis in UMMs","display_name":"Calibrating Attention-Based Token Pruning for Efficient Multi-Image Synthesis in UMMs","publication_year":2026,"publication_date":"2026-10-29","ids":{"openalex":"https://openalex.org/W7224895694","doi":"https://doi.org/10.48550/arxiv.2610.39467"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.39467","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.39467","pdf_url":"https://arxiv.org/pdf/2610.39467","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","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2610.39467","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5160223653","display_name":"Zimeng Wu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I82880672","display_name":"Beihang University","ror":"https://ror.org/00wk2mp56","country_code":"CN","type":"education","lineage":["https://openalex.org/I82880672"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wu, Zimeng","raw_affiliation_strings":["Beihang University","KlingAI Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Beihang University","institution_ids":["https://openalex.org/I82880672"]},{"raw_affiliation_string":"KlingAI Research","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101602703","display_name":"Qijian Tian","orcid":"https://orcid.org/0000-0001-7218-4978","observed_orcids":["https://orcid.org/0000-0001-7218-4978"]},"institutions":[{"id":"https://openalex.org/I183067930","display_name":"Shanghai Jiao Tong University","ror":"https://ror.org/0220qvk04","country_code":"CN","type":"education","lineage":["https://openalex.org/I183067930"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Tian, Qijian","raw_affiliation_strings":["Shanghai Jiao Tong University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Shanghai Jiao Tong University","institution_ids":["https://openalex.org/I183067930"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133199901","display_name":"Penghui Du","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Du, Penghui","raw_affiliation_strings":["KlingAI Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"KlingAI Research","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5108609473","display_name":"Changqian Yu","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yu, Changqian","raw_affiliation_strings":["KlingAI Research"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"KlingAI Research","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5160191169","display_name":"Jiaxin Chen","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I82880672","display_name":"Beihang University","ror":"https://ror.org/00wk2mp56","country_code":"CN","type":"education","lineage":["https://openalex.org/I82880672"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Chen, Jiaxin","raw_affiliation_strings":["Beihang University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Beihang University","institution_ids":["https://openalex.org/I82880672"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5160237500","display_name":"Yunhong Wang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I82880672","display_name":"Beihang University","ror":"https://ror.org/00wk2mp56","country_code":"CN","type":"education","lineage":["https://openalex.org/I82880672"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Wang, Yunhong","raw_affiliation_strings":["Beihang University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Beihang University","institution_ids":["https://openalex.org/I82880672"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.9824000000953674,"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":0.9824000000953674,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.008799999952316284,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.007300000172108412,"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/token-pruning","display_name":"token pruning","score":0.875},{"id":"https://openalex.org/keywords/unified-multimodal-models","display_name":"unified multimodal models","score":0.9160000085830688},{"id":"https://openalex.org/keywords/context-compression","display_name":"context compression","score":0.7680000066757202},{"id":"https://openalex.org/keywords/inference-acceleration","display_name":"inference acceleration","score":0.5809999704360962}],"concepts":[{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.8453999757766724},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7146999835968018},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.578499972820282},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5703999996185303},{"id":"https://openalex.org/C188441871","wikidata":"https://www.wikidata.org/wiki/Q7554146","display_name":"Softmax function","level":3,"score":0.5472999811172485},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.5152999758720398},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4781999886035919},{"id":"https://openalex.org/C165838908","wikidata":"https://www.wikidata.org/wiki/Q736777","display_name":"Calibration","level":2,"score":0.4627000093460083},{"id":"https://openalex.org/C189430467","wikidata":"https://www.wikidata.org/wiki/Q7293293","display_name":"Ranking (information retrieval)","level":2,"score":0.454800009727478},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.42260000109672546},{"id":"https://openalex.org/C17020691","wikidata":"https://www.wikidata.org/wiki/Q139677","display_name":"Operator (biology)","level":5,"score":0.4065000116825104},{"id":"https://openalex.org/C2780522230","wikidata":"https://www.wikidata.org/wiki/Q1140419","display_name":"Ambiguity","level":2,"score":0.38659998774528503},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3709999918937683},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.3215999901294708},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.30820000171661377},{"id":"https://openalex.org/C42058472","wikidata":"https://www.wikidata.org/wiki/Q810214","display_name":"Base (topology)","level":2,"score":0.30730000138282776},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.3068000078201294},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.30559998750686646},{"id":"https://openalex.org/C115961682","wikidata":"https://www.wikidata.org/wiki/Q860623","display_name":"Image (mathematics)","level":2,"score":0.29339998960494995},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2775999903678894},{"id":"https://openalex.org/C2780898871","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Performance metric","level":2,"score":0.26499998569488525},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.26409998536109924},{"id":"https://openalex.org/C124304363","wikidata":"https://www.wikidata.org/wiki/Q673661","display_name":"Abstraction","level":2,"score":0.2605000138282776},{"id":"https://openalex.org/C31170391","wikidata":"https://www.wikidata.org/wiki/Q188619","display_name":"Hierarchy","level":2,"score":0.2535000145435333},{"id":"https://openalex.org/C2780009758","wikidata":"https://www.wikidata.org/wiki/Q6804172","display_name":"Measure (data warehouse)","level":2,"score":0.25099998712539673}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.39467","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.39467","pdf_url":"https://arxiv.org/pdf/2610.39467","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.2610.39467","doi":"https://doi.org/10.48550/arxiv.2610.39467","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.39467","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:2610.39467","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.39467","pdf_url":"https://arxiv.org/pdf/2610.39467","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/W7224895694.pdf","grobid_xml":"https://content.openalex.org/works/W7224895694.grobid-xml"},"referenced_works_count":13,"referenced_works":["https://openalex.org/W4407794264","https://openalex.org/W4415335344","https://openalex.org/W4416010170","https://openalex.org/W4416780398","https://openalex.org/W6948058629","https://openalex.org/W6948505238","https://openalex.org/W7109105434","https://openalex.org/W7125269651","https://openalex.org/W7127370116","https://openalex.org/W7141258663","https://openalex.org/W7154380296","https://openalex.org/W7160991674","https://openalex.org/W7169525321"],"related_works":["https://openalex.org/W7161051236","https://openalex.org/W7135202982","https://openalex.org/W4407359167","https://openalex.org/W7204940474","https://openalex.org/W4412082472","https://openalex.org/W4415194422","https://openalex.org/W7138945373","https://openalex.org/W4407384793","https://openalex.org/W4417357134","https://openalex.org/W4414820693"],"abstract_inverted_index":{"Unified":[0],"Multimodal":[1],"Models":[2],"(UMMs)":[3],"enable":[4],"powerful":[5],"multi-image":[6],"synthesis":[7],"but":[8],"incur":[9],"substantial":[10],"computation":[11],"from":[12,31,35],"long":[13],"visual":[14],"contexts.":[15],"While":[16],"token":[17],"pruning":[18,84,94,131],"offers":[19],"a":[20,100,107,117],"direct":[21],"route":[22],"to":[23,63,92,129,150],"accelerating":[24],"inference,":[25],"existing":[26],"methods":[27],"are":[28],"primarily":[29],"adapted":[30],"understanding":[32],"tasks,":[33],"suffering":[34],"insufficient":[36],"reference":[37],"image":[38],"exploration,":[39],"confused":[40],"conditional":[41],"guidance,":[42],"and":[43,99,138],"pooling-induced":[44,113],"ambiguity":[45],"that":[46,105,141],"degrades":[47],"generation":[48],"quality.":[49],"To":[50],"address":[51],"these":[52],"issues,":[53],"we":[54,87],"propose":[55],"CATP,":[56],"the":[57,68,123],"first":[58],"training-free":[59],"context-compression":[60],"method":[61],"tailored":[62],"UMM":[64],"generation,":[65],"which":[66],"\\textbf{C}alibrates":[67],"discrepancies":[69],"of":[70],"general":[71],"\\textbf{A}ttention-based":[72],"\\textbf{T}oken":[73],"\\textbf{P}runing":[74],"for":[75,112],"better":[76],"quality-efficiency":[77],"trade-off.":[78],"Concretely,":[79],"building":[80],"on":[81],"independent":[82],"layer-wise":[83],"during":[85],"denoising,":[86],"introduce":[88],"Guidance-Consistent":[89],"Pruning":[90],"(GCP)":[91],"synchronize":[93],"decisions":[95],"across":[96,135],"conditioning":[97],"branches,":[98],"Taylor-Calibrated":[101],"Attention":[102],"Metric":[103],"(TCAM)":[104],"derives":[106],"theoretically":[108],"grounded":[109],"second-order":[110],"correction":[111],"bias.":[114],"In":[115],"addition,":[116],"Fast":[118],"Scoring":[119],"Operator":[120],"(FSO)":[121],"fuses":[122],"calibration":[124],"into":[125],"tiled":[126],"online":[127],"Softmax":[128],"reduce":[130],"cost.":[132],"Extensive":[133],"experiments":[134],"representative":[136],"UMMs":[137],"benchmarks":[139],"demonstrate":[140],"CATP":[142],"consistently":[143],"outperforms":[144],"state-of-the-art":[145],"token-reduction":[146],"methods,":[147],"achieving":[148],"up":[149],"2.09$\\times$":[151],"attention-core":[152],"speedup.":[153]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-31T00:00:00"}