{"id":"https://openalex.org/W7165190660","doi":"https://doi.org/10.48550/arxiv.2606.19150","title":"Complementary Attention Head Pruning for Efficient Transformers","display_name":"Complementary Attention Head Pruning for Efficient Transformers","publication_year":2026,"publication_date":"2026-06-17","ids":{"openalex":"https://openalex.org/W7165190660","doi":"https://doi.org/10.48550/arxiv.2606.19150"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2606.19150","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.19150","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,"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":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2606.19150","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5138902685","display_name":"Yaniv Livertovsky","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Livertovsky, Yaniv","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5053754779","display_name":"Shahar Somin","orcid":"https://orcid.org/0009-0007-2400-9068"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Somin, Shahar","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5077171115","display_name":"Gonen Singer","orcid":"https://orcid.org/0000-0002-2610-9579"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Singer, Gonen","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":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.3903000056743622,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.3903000056743622,"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":0.11980000138282776,"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/T10775","display_name":"Generative Adversarial Networks and Image Synthesis","score":0.11209999769926071,"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/pruning","display_name":"Pruning","score":0.6983000040054321},{"id":"https://openalex.org/keywords/hyperparameter","display_name":"Hyperparameter","score":0.6292999982833862},{"id":"https://openalex.org/keywords/cluster-analysis","display_name":"Cluster analysis","score":0.5525000095367432},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.48989999294281006},{"id":"https://openalex.org/keywords/ranking","display_name":"Ranking (information retrieval)","score":0.4260999858379364},{"id":"https://openalex.org/keywords/set","display_name":"Set (abstract data type)","score":0.3846000134944916},{"id":"https://openalex.org/keywords/selection","display_name":"Selection (genetic algorithm)","score":0.37389999628067017},{"id":"https://openalex.org/keywords/pattern-recognition","display_name":"Pattern recognition (psychology)","score":0.31040000915527344}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7027000188827515},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.6983000040054321},{"id":"https://openalex.org/C8642999","wikidata":"https://www.wikidata.org/wiki/Q4171168","display_name":"Hyperparameter","level":2,"score":0.6292999982833862},{"id":"https://openalex.org/C73555534","wikidata":"https://www.wikidata.org/wiki/Q622825","display_name":"Cluster analysis","level":2,"score":0.5525000095367432},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.54830002784729},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.5142999887466431},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.48989999294281006},{"id":"https://openalex.org/C189430467","wikidata":"https://www.wikidata.org/wiki/Q7293293","display_name":"Ranking (information retrieval)","level":2,"score":0.4260999858379364},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.38530001044273376},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.3846000134944916},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.37389999628067017},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.31040000915527344},{"id":"https://openalex.org/C2780312720","wikidata":"https://www.wikidata.org/wiki/Q5689100","display_name":"Head (geology)","level":2,"score":0.30640000104904175},{"id":"https://openalex.org/C84525736","wikidata":"https://www.wikidata.org/wiki/Q831366","display_name":"Decision tree","level":2,"score":0.2964000105857849},{"id":"https://openalex.org/C79581498","wikidata":"https://www.wikidata.org/wiki/Q1367530","display_name":"Suite","level":2,"score":0.29440000653266907},{"id":"https://openalex.org/C30684385","wikidata":"https://www.wikidata.org/wiki/Q176509","display_name":"Ringing","level":3,"score":0.2712000012397766},{"id":"https://openalex.org/C90119067","wikidata":"https://www.wikidata.org/wiki/Q43260","display_name":"Polynomial","level":2,"score":0.27059999108314514},{"id":"https://openalex.org/C205711294","wikidata":"https://www.wikidata.org/wiki/Q176953","display_name":"Rendering (computer graphics)","level":2,"score":0.25870001316070557},{"id":"https://openalex.org/C34736171","wikidata":"https://www.wikidata.org/wiki/Q918333","display_name":"Preprocessor","level":2,"score":0.25540000200271606},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.2549000084400177},{"id":"https://openalex.org/C105339364","wikidata":"https://www.wikidata.org/wiki/Q2297740","display_name":"Software deployment","level":2,"score":0.2547999918460846}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2606.19150","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.19150","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,"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":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2606.19150","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2606.19150","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,"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":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"The":[0],"remarkable":[1],"success":[2],"of":[3,20,108,127,193,216],"Transformer-based":[4],"models":[5],"in":[6,25,89,149,179,203,219],"natural":[7],"language":[8],"processing":[9],"stems":[10],"from":[11,50],"architectural":[12],"scaling,":[13],"which":[14,48,197],"leads":[15,144],"to":[16,34,100,145,199,206],"a":[17,32,72,81,104,114,135,146,212],"large":[18],"number":[19,126],"parameters":[21],"and":[22,54,102,163,209],"hinders":[23],"deployment":[24],"resource-constrained":[26],"environments.":[27],"While":[28],"structured":[29],"pruning":[30,119,141,195],"offers":[31],"pathway":[33],"compression,":[35],"existing":[36],"state-of-the-art":[37],"methods":[38],"often":[39],"rely":[40],"on":[41,160],"gradient-based":[42,194],"importance":[43],"ranking":[44],"or":[45,118],"stochastic":[46],"gating,":[47],"suffer":[49],"instability,":[51],"structural":[52,184],"degeneration,":[53],"the":[55,121,125,154,161,190,207,220],"need":[56],"for":[57],"extensive":[58],"manual":[59],"hyperparameter":[60],"tuning.":[61],"In":[62],"this":[63],"paper,":[64],"we":[65],"introduce":[66],"CAHP":[67,91,173,188],"(Complementary":[68],"Attention":[69],"Head":[70],"Pruning),":[71],"novel":[73],"post-hoc":[74],"framework":[75,122],"that":[76,172,187],"redefines":[77],"head":[78],"selection":[79],"as":[80,151],"global":[82],"graph-theoretical":[83],"problem.":[84],"Rather":[85],"than":[86],"evaluating":[87],"heads":[88,130,143,201,218],"isolation,":[90],"utilizes":[92],"graph-based":[93],"clustering":[94],"combined":[95],"with":[96],"information-theoretic":[97],"distance":[98],"measures":[99],"identify":[101],"preserve":[103,200],"topologically":[105],"diverse":[106],"subset":[107],"complementary":[109],"attention":[110,129,217],"heads.":[111],"Without":[112],"requiring":[113],"predefined":[115],"sparsity":[116],"level":[117],"ratio,":[120],"automatically":[123],"determines":[124],"selected":[128],"across":[131,166],"layers":[132,204],"by":[133,153],"identifying":[134],"diminishing":[136],"marginal":[137],"performance":[138],"curve,":[139],"where":[140],"additional":[142],"sharp":[147],"degradation":[148],"performance,":[150],"determined":[152],"chosen":[155],"polynomial":[156],"degree.":[157],"Extensive":[158],"evaluations":[159],"SST-5":[162],"MNLI":[164],"benchmarks,":[165],"different":[167],"Transformer":[168],"model":[169],"scales,":[170],"demonstrate":[171],"consistently":[174],"outperforms":[175],"competitive":[176],"baselines,":[177],"particularly":[178],"high-compression":[180],"regimes.":[181],"Furthermore,":[182],"our":[183],"analysis":[185],"shows":[186],"avoids":[189],"\"proximity":[191],"bias\"":[192],"methods,":[196],"tend":[198],"mainly":[202],"close":[205],"output,":[208],"instead":[210],"retains":[211],"functionally":[213],"critical":[214],"set":[215],"model's":[221],"intermediate":[222],"layers.":[223]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-06-19T00:00:00"}