{"id":"https://openalex.org/W4408348458","doi":"https://doi.org/10.20944/preprints202503.0590.v1","title":"Token-Level Pruning in Attention Models","display_name":"Token-Level Pruning in Attention Models","publication_year":2025,"publication_date":"2025-03-10","ids":{"openalex":"https://openalex.org/W4408348458","doi":"https://doi.org/10.20944/preprints202503.0590.v1"},"language":"en","primary_location":{"id":"doi:10.20944/preprints202503.0590.v1","doi":"https://doi.org/10.20944/preprints202503.0590.v1","is_oa":true,"landing_page_url":"https://doi.org/10.20944/preprints202503.0590.v1","pdf_url":null,"source":{"id":"https://openalex.org/S6309402219","display_name":"Preprints.org","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":["https://openalex.org/P4310310987"],"host_organization_lineage_names":["Multidisciplinary Digital Publishing Institute"],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":null,"raw_type":"posted-content"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.20944/preprints202503.0590.v1","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5116592324","display_name":"Shui Xiuying","orcid":"https://orcid.org/0009-0003-8139-3714","observed_orcids":["https://orcid.org/0009-0003-8139-3714"]},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Shui Xiuying","raw_affiliation_strings":[],"raw_orcid":"https://orcid.org/0009-0003-8139-3714","affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":["https://openalex.org/A5116592324"],"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":0.6064000129699707,"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":0.6064000129699707,"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":0.32179999351501465,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.054099999368190765,"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/token-pruning","display_name":"token pruning","score":0.8379999995231628},{"id":"https://openalex.org/keywords/attention-models","display_name":"attention-based models","score":0.4560000002384186},{"id":"https://openalex.org/keywords/transformer","display_name":"transformer models","score":0.7049999833106995},{"id":"https://openalex.org/keywords/inference-efficiency","display_name":"inference efficiency","score":0.8550000190734863},{"id":"https://openalex.org/keywords/resource-constrained-deployment","display_name":"resource-constrained deployment","score":0.8140000104904175},{"id":"https://openalex.org/keywords/distribution-shift-robustness","display_name":"distribution shift robustness","score":0.75},{"id":"https://openalex.org/keywords/model-compression","display_name":"model compression","score":0.4819999933242798}],"concepts":[{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.752090334892273},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.5987251400947571},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4480368494987488},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.34206074476242065},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.34017473459243774},{"id":"https://openalex.org/C38652104","wikidata":"https://www.wikidata.org/wiki/Q3510521","display_name":"Computer security","level":1,"score":0.09969314932823181},{"id":"https://openalex.org/C144027150","wikidata":"https://www.wikidata.org/wiki/Q48803","display_name":"Horticulture","level":1,"score":0.049794554710388184},{"id":"https://openalex.org/C86803240","wikidata":"https://www.wikidata.org/wiki/Q420","display_name":"Biology","level":0,"score":0.04831990599632263}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.20944/preprints202503.0590.v1","doi":"https://doi.org/10.20944/preprints202503.0590.v1","is_oa":true,"landing_page_url":"https://doi.org/10.20944/preprints202503.0590.v1","pdf_url":null,"source":{"id":"https://openalex.org/S6309402219","display_name":"Preprints.org","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":["https://openalex.org/P4310310987"],"host_organization_lineage_names":["Multidisciplinary Digital Publishing Institute"],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":null,"raw_type":"posted-content"}],"best_oa_location":{"id":"doi:10.20944/preprints202503.0590.v1","doi":"https://doi.org/10.20944/preprints202503.0590.v1","is_oa":true,"landing_page_url":"https://doi.org/10.20944/preprints202503.0590.v1","pdf_url":null,"source":{"id":"https://openalex.org/S6309402219","display_name":"Preprints.org","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":["https://openalex.org/P4310310987"],"host_organization_lineage_names":["Multidisciplinary Digital Publishing Institute"],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"acceptedVersion","is_accepted":true,"is_published":false,"raw_source_name":null,"raw_type":"posted-content"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4408450891","https://openalex.org/W4408756850","https://openalex.org/W4408138862","https://openalex.org/W4408007712","https://openalex.org/W4408826259","https://openalex.org/W4408022330","https://openalex.org/W3180037928","https://openalex.org/W4382322807","https://openalex.org/W7217942532","https://openalex.org/W7127923142"],"abstract_inverted_index":{"Transformer-based":[0],"models":[1,201],"have":[2],"revolutionized":[3],"natural":[4],"language":[5],"processing":[6],"(NLP),":[7],"achieving":[8],"state-of-the-art":[9],"performance":[10],"across":[11,96],"a":[12,68],"wide":[13],"range":[14],"of":[15,47,71],"tasks.":[16],"However,":[17],"their":[18,102],"high":[19],"computational":[20,59],"cost":[21],"and":[22,82,100,108,127,158,167,180,195],"memory":[23],"requirements":[24],"pose":[25],"significant":[26],"challenges":[27,128,166],"for":[28,115,184,193,202],"real-world":[29,120],"deployment,":[30],"particularly":[31],"in":[32,119],"resource-constrained":[33],"environments.":[34],"Token":[35],"pruning":[36,73,118,161],"has":[37],"emerged":[38],"as":[39,147],"an":[40],"effective":[41],"technique":[42],"to":[43,130,150,172,189,198],"enhance":[44],"the":[45,87,134,174],"efficiency":[46,203],"transformers":[48],"by":[49,138],"dynamically":[50],"removing":[51],"less":[52],"informative":[53],"tokens":[54],"during":[55],"inference,":[56],"thereby":[57],"reducing":[58],"complexity":[60],"while":[61],"maintaining":[62],"competitive":[63],"accuracy.":[64],"This":[65],"survey":[66,187],"provides":[67],"comprehensive":[69],"review":[70],"token":[72,117,139],"methods,":[74],"categorizing":[75],"them":[76],"into":[77],"attention-based,":[78],"gradient-based,":[79],"reinforcement":[80],"learning-based,":[81],"hybrid":[83],"approaches.":[84],"We":[85,163],"analyze":[86],"theoretical":[88],"foundations":[89],"behind":[90],"these":[91,165],"techniques,":[92],"discuss":[93],"empirical":[94],"evaluations":[95],"various":[97],"NLP":[98],"benchmarks,":[99],"explore":[101],"impact":[103],"on":[104],"model":[105],"accuracy,":[106],"efficiency,":[107],"generalization.":[109],"Additionally,":[110],"we":[111],"examine":[112],"practical":[113],"considerations":[114],"implementing":[116],"applications,":[121],"including":[122],"optimization":[123],"strategies,":[124],"hardware":[125],"compatibility,":[126],"related":[129],"dynamic":[131],"execution.":[132],"Despite":[133],"promising":[135],"results":[136],"achieved":[137],"pruning,":[140],"several":[141],"open":[142],"research":[143,170],"questions":[144],"remain,":[145],"such":[146],"improving":[148],"adaptability":[149],"different":[151],"tasks,":[152],"ensuring":[153],"robustness":[154],"under":[155],"distribution":[156],"shifts,":[157],"developing":[159],"hardware-aware":[160],"techniques.":[162],"highlight":[164],"outline":[168],"future":[169],"directions":[171],"advance":[173],"field.":[175],"By":[176],"consolidating":[177],"existing":[178],"knowledge":[179],"identifying":[181],"key":[182],"areas":[183],"innovation,":[185],"this":[186],"aims":[188],"provide":[190],"valuable":[191],"insights":[192],"researchers":[194],"practitioners":[196],"seeking":[197],"optimize":[199],"transformer-based":[200],"without":[204],"sacrificing":[205],"performance.":[206]},"counts_by_year":[],"updated_date":"2026-09-21T07:43:02.447757","created_date":"2025-03-13T00:00:00"}