{"id":"https://openalex.org/W7166841818","doi":"https://doi.org/10.18653/v1/2026.acl-long.1877","title":"BOSCH: Black-Box Binary Optimization for Short-Context Attention-Head Selection in LLMs","display_name":"BOSCH: Black-Box Binary Optimization for Short-Context Attention-Head Selection in LLMs","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166841818","doi":"https://doi.org/10.18653/v1/2026.acl-long.1877"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.acl-long.1877","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.1877","pdf_url":"https://aclanthology.org/2026.acl-long.1877.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.1877.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5139747159","display_name":"Abbas Ghaddar","orcid":null},"institutions":[{"id":"https://openalex.org/I4210159102","display_name":"Huawei Technologies (Sweden)","ror":"https://ror.org/0500fyd17","country_code":"SE","type":"company","lineage":["https://openalex.org/I2250955327","https://openalex.org/I4210159102"]}],"countries":["SE"],"is_corresponding":false,"raw_author_name":"Abbas Ghaddar","raw_affiliation_strings":["Huawei Noah's Ark Lab , Montreal Research Center , Canada"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Huawei Noah's Ark Lab , Montreal Research Center , Canada","institution_ids":["https://openalex.org/I4210159102"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5008923192","display_name":"Ivan Kobyzev","orcid":"https://orcid.org/0000-0003-1934-4842"},"institutions":[{"id":"https://openalex.org/I4210159102","display_name":"Huawei Technologies (Sweden)","ror":"https://ror.org/0500fyd17","country_code":"SE","type":"company","lineage":["https://openalex.org/I2250955327","https://openalex.org/I4210159102"]}],"countries":["SE"],"is_corresponding":false,"raw_author_name":"Ivan Kobyzev","raw_affiliation_strings":["Huawei Noah's Ark Lab , Montreal Research Center , Canada"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Huawei Noah's Ark Lab , Montreal Research Center , Canada","institution_ids":["https://openalex.org/I4210159102"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5019715118","display_name":"Boxing Chen","orcid":"https://orcid.org/0000-0002-3170-4858"},"institutions":[{"id":"https://openalex.org/I4210159102","display_name":"Huawei Technologies (Sweden)","ror":"https://ror.org/0500fyd17","country_code":"SE","type":"company","lineage":["https://openalex.org/I2250955327","https://openalex.org/I4210159102"]}],"countries":["SE"],"is_corresponding":false,"raw_author_name":"Boxing Chen","raw_affiliation_strings":["Huawei Noah's Ark Lab , Montreal Research Center , Canada"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Huawei Noah's Ark Lab , Montreal Research Center , Canada","institution_ids":["https://openalex.org/I4210159102"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5139763720","display_name":"Yufei Cui","orcid":null},"institutions":[{"id":"https://openalex.org/I4210159102","display_name":"Huawei Technologies (Sweden)","ror":"https://ror.org/0500fyd17","country_code":"SE","type":"company","lineage":["https://openalex.org/I2250955327","https://openalex.org/I4210159102"]}],"countries":["SE"],"is_corresponding":false,"raw_author_name":"Yufei Cui","raw_affiliation_strings":["Huawei Noah's Ark Lab , Montreal Research Center , Canada"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Huawei Noah's Ark Lab , Montreal Research Center , Canada","institution_ids":["https://openalex.org/I4210159102"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I4210159102"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.83163102,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"40417","last_page":"40444"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T12535","display_name":"Machine Learning and Data Classification","score":0.13410000503063202,"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/T12535","display_name":"Machine Learning and Data Classification","score":0.13410000503063202,"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/T10203","display_name":"Recommender Systems and Techniques","score":0.09319999814033508,"subfield":{"id":"https://openalex.org/subfields/1710","display_name":"Information Systems"},"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":0.07739999890327454,"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/selection","display_name":"Selection (genetic algorithm)","score":0.5370000004768372},{"id":"https://openalex.org/keywords/binary-number","display_name":"Binary number","score":0.34369999170303345},{"id":"https://openalex.org/keywords/binary-data","display_name":"Binary data","score":0.3111000061035156},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.27309998869895935},{"id":"https://openalex.org/keywords/genetic-algorithm","display_name":"Genetic algorithm","score":0.2687000036239624}],"concepts":[{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.5370000004768372},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4821999967098236},{"id":"https://openalex.org/C48372109","wikidata":"https://www.wikidata.org/wiki/Q3913","display_name":"Binary number","level":2,"score":0.34369999170303345},{"id":"https://openalex.org/C2779190172","wikidata":"https://www.wikidata.org/wiki/Q4913888","display_name":"Binary data","level":3,"score":0.3111000061035156},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.30090001225471497},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2948000133037567},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.27309998869895935},{"id":"https://openalex.org/C8880873","wikidata":"https://www.wikidata.org/wiki/Q187787","display_name":"Genetic algorithm","level":2,"score":0.2687000036239624},{"id":"https://openalex.org/C147764199","wikidata":"https://www.wikidata.org/wiki/Q6865248","display_name":"Minification","level":2,"score":0.258899986743927},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.2565000057220459},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.2517000138759613}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.acl-long.1877","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.1877","pdf_url":"https://aclanthology.org/2026.acl-long.1877.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.1877","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.acl-long.1877","pdf_url":"https://aclanthology.org/2026.acl-long.1877.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":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166841818.pdf","grobid_xml":"https://content.openalex.org/works/W7166841818.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Post-training":[0],"hybridization":[1,23],"of":[2,175,191],"large":[3],"language":[4],"models":[5],"(LLMs)":[6],"often":[7],"replaces":[8],"quadratic":[9],"selfattention":[10],"with":[11,155],"sliding-window":[12],"attention":[13],"(SWA)":[14],"to":[15,45,135,171],"reduce":[16],"KV":[17],"cache":[18],"usage":[19],"and":[20,51,98,120,149,170],"improve":[21],"latency.Existing":[22],"schemes":[24,47],"are":[25,54],"typically":[26],"defined":[27],"either":[28],"at":[29,36,158],"the":[30,37,59,91,176,189],"layer":[31],"level":[32,39],"(e.g.,":[33],"interleaving)":[34],"or":[35],"head":[38],"via":[40,107],"static":[41,63,152],"rankings":[42,65],"from":[43,67,133],"local":[44,50],"global.Layer-level":[46],"ignore":[48],"that":[49,89,143],"global":[52],"dependencies":[53],"routed":[55],"through":[56],"heads":[57,178],"within":[58,125],"same":[60],"layer,":[61],"while":[62],"head-level":[64,153,193],"suffer":[66],"entanglement:":[68],"a":[69,86,94,172],"head's":[70],"local/global":[71],"behavior":[72],"can":[73],"change":[74],"after":[75],"hybridization.We":[76],"propose":[77],"BOSCH,":[78],"Black-box":[79],"Binary":[80],"Optimization":[81],"for":[82,182,195],"Short-context":[83],"Head":[84],"Selection,":[85],"training-free":[87],"method":[88],"formulates":[90],"problem":[92],"as":[93],"Large":[95],"Neighborhood":[96],"Search":[97],"decomposes":[99],"it":[100],"into":[101],"three":[102],"subproblems:":[103],"(i)":[104],"layer-importance":[105],"detection":[106],"small-budget":[108],"black-box":[109],"probes,":[110],"(ii)":[111],"adaptive":[112],"per-layer":[113],"SWA-ratio":[114],"assignment":[115],"based":[116],"on":[117,129,202],"these":[118],"sensitivities,":[119],"(iii)":[121],"grouped":[122],"headlevel":[123],"optimization":[124],"ratio":[126,198],"buckets.Extensive":[127],"experiments":[128],"4":[130,139],"LLMs":[131],"ranging":[132],"1.7B":[134],"30B":[136],"parameters,":[137],"across":[138,184],"SWA":[140,160,186],"ratios,":[141,187],"show":[142],"BOSCH":[144,164,183],"consistently":[145],"outperforms":[146],"layerlevel":[147],"heuristics":[148],"6":[150],"strong":[151],"methods,":[154],"larger":[156],"gains":[157],"higher":[159,173],"ratios.Under":[161],"continual":[162],"pretraining,":[163],"recover":[165],"original":[166],"long-context":[167],"performance":[168],"faster":[169],"level.Analysis":[174],"selected":[177],"reveals":[179],"substantial":[180],"turnover":[181],"different":[185],"underscoring":[188],"importance":[190],"performing":[192],"selection":[194],"each":[196],"target":[197],"rather":[199],"than":[200],"relying":[201],"fixed":[203],"locality":[204],"rankings.":[205]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}