{"id":"https://openalex.org/W7166815409","doi":"https://doi.org/10.18653/v1/2026.acl-long.450","title":"Sparrow: Text-Anchored Window Attention with Visual-Semantic Glimpsing for Speculative Decoding in Video LLMs","display_name":"Sparrow: Text-Anchored Window Attention with Visual-Semantic Glimpsing for Speculative Decoding in Video LLMs","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166815409","doi":"https://doi.org/10.18653/v1/2026.acl-long.450"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.acl-long.450","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.450","pdf_url":"https://aclanthology.org/2026.acl-long.450.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.acl-long.450.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5139835622","display_name":"Libo Zhang","orcid":null},"institutions":[{"id":"https://openalex.org/I170215575","display_name":"National University of Defense Technology","ror":"https://ror.org/05d2yfz11","country_code":"CN","type":"education","lineage":["https://openalex.org/I170215575"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Libo Zhang","raw_affiliation_strings":["National Key Laboratory of Parallel and Distributed Computing College of Computer Science and Technology National University of Defense Technology , Changsha , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"National Key Laboratory of Parallel and Distributed Computing College of Computer Science and Technology National University of Defense Technology , Changsha , China","institution_ids":["https://openalex.org/I170215575"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139830114","display_name":"Zhaoning Zhang","orcid":null},"institutions":[{"id":"https://openalex.org/I170215575","display_name":"National University of Defense Technology","ror":"https://ror.org/05d2yfz11","country_code":"CN","type":"education","lineage":["https://openalex.org/I170215575"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Zhaoning Zhang","raw_affiliation_strings":["National Key Laboratory of Parallel and Distributed Computing College of Computer Science and Technology National University of Defense Technology , Changsha , China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"National Key Laboratory of Parallel and Distributed Computing College of Computer Science and Technology National University of Defense Technology , Changsha , China","institution_ids":["https://openalex.org/I170215575"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139828122","display_name":"Hongwanyang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hongwanyang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5134187939","display_name":"Peng Qiao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Peng Qiao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5139736778","display_name":"Dongsheng Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dongsheng Li","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.86162558,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"9928","last_page":"9942"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.6690000295639038,"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":0.6690000295639038,"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/T10028","display_name":"Topic Modeling","score":0.05429999902844429,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.05009999871253967,"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/decoding-methods","display_name":"Decoding methods","score":0.4788999855518341},{"id":"https://openalex.org/keywords/window","display_name":"Window (computing)","score":0.3643999993801117},{"id":"https://openalex.org/keywords/key","display_name":"Key (lock)","score":0.30979999899864197},{"id":"https://openalex.org/keywords/scheme","display_name":"Scheme (mathematics)","score":0.24639999866485596},{"id":"https://openalex.org/keywords/encoding","display_name":"Encoding (memory)","score":0.24169999361038208}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6409000158309937},{"id":"https://openalex.org/C57273362","wikidata":"https://www.wikidata.org/wiki/Q576722","display_name":"Decoding methods","level":2,"score":0.4788999855518341},{"id":"https://openalex.org/C2778751112","wikidata":"https://www.wikidata.org/wiki/Q835016","display_name":"Window (computing)","level":2,"score":0.3643999993801117},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.31279999017715454},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.30979999899864197},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.27399998903274536},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.24639999866485596},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.2451000064611435},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.24169999361038208},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.22669999301433563}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.acl-long.450","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.450","pdf_url":"https://aclanthology.org/2026.acl-long.450.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.acl-long.450","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.450","pdf_url":"https://aclanthology.org/2026.acl-long.450.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":"Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G2277931485","display_name":null,"funder_award_id":"62421002","funder_id":"https://openalex.org/F4320321001","funder_display_name":"National Natural Science Foundation of China"},{"id":"https://openalex.org/G2498301691","display_name":null,"funder_award_id":"JYB2025XDXM202","funder_id":"https://openalex.org/F4320321106","funder_display_name":"Ministry of Education of the People's Republic of China"}],"funders":[{"id":"https://openalex.org/F4320321001","display_name":"National Natural Science Foundation of China","ror":"https://ror.org/01h0zpd94"},{"id":"https://openalex.org/F4320321106","display_name":"Ministry of Education of the People's Republic of China","ror":"https://ror.org/01mv9t934"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166815409.pdf","grobid_xml":"https://content.openalex.org/works/W7166815409.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Although":[0],"speculative":[1],"decoding":[2],"is":[3,134],"widely":[4],"used":[5],"to":[6,19,40,100,105,115,136],"accelerate":[7],"Vision-Language":[8],"Models":[9,23],"(VLMs)":[10],"inference,":[11],"it":[12],"faces":[13],"severe":[14],"performance":[15,159],"collapse":[16],"when":[17],"applied":[18],"Video":[20],"Large":[21],"Language":[22],"(Vid-LLMs).The":[24],"draft":[25,118],"model":[26,119],"typically":[27],"falls":[28],"into":[29,64],"the":[30,85,106,117,138,158],"trap":[31],"of":[32,149],"attention":[33,95],"dilution":[34],"and":[35,44,109,164],"negative":[36],"visual":[37,50,59,74,103,112,128,154],"gain":[38],"due":[39],"key-value":[41],"cache":[42],"explosion":[43],"context":[45],"window":[46,94],"mismatches.We":[47],"observe":[48],"a":[49,130,166],"semantic":[51],"internalization":[52],"phenomenon":[53],"in":[54,161],"Vid-LLMs,":[55],"indicating":[56],"that":[57,143],"critical":[58],"semantics":[60],"are":[61],"implicitly":[62],"encoded":[63],"text":[65],"hidden":[66,97],"states":[67],"during":[68,78],"deep-layer":[69],"interactions,":[70],"which":[71,89],"renders":[72],"raw":[73],"inputs":[75],"structurally":[76],"redundant":[77],"deep":[79],"inference.To":[80],"address":[81],"this,":[82],"we":[83],"propose":[84],"Sparrow":[86,144],"1":[87],"framework,":[88],"first":[90],"utilizes":[91],"visually-aware":[92],"text-anchored":[93],"via":[96],"state":[98,113],"reuse":[99],"fully":[101],"offload":[102],"computation":[104],"target":[107],"model,":[108],"leverages":[110],"intermediate-layer":[111],"bridging":[114],"train":[116],"with":[120,152],"semantic-rich":[121],"intermediate":[122],"states,":[123],"thereby":[124],"filtering":[125],"out":[126],"lowlevel":[127],"noise.Additionally,":[129],"multi-token":[131],"prediction":[132],"strategy":[133],"introduced":[135],"bridge":[137],"training-inference":[139],"distribution":[140],"shift.Experiments":[141],"show":[142],"achieves":[145],"an":[146],"average":[147],"speedup":[148],"2.82\u00d7":[150],"even":[151],"25k":[153],"tokens,":[155],"effectively":[156],"resolving":[157],"degradation":[160],"long":[162,171],"sequences":[163],"offering":[165],"practical":[167],"solution":[168],"for":[169],"real-time":[170],"video":[172],"tasks.":[173]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}