{"id":"https://openalex.org/W7162400291","doi":"https://doi.org/10.48550/arxiv.2605.24168","title":"Inference Time Context Sparsity: Illusion or Opportunity?","display_name":"Inference Time Context Sparsity: Illusion or Opportunity?","publication_year":2026,"publication_date":"2026-05-22","ids":{"openalex":"https://openalex.org/W7162400291","doi":"https://doi.org/10.48550/arxiv.2605.24168"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.24168","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.24168","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":"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"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.24168","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137067935","display_name":"Sahil Joshi","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Joshi, Sahil","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137057142","display_name":"Prithvi Dixit","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Dixit, Prithvi","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5020686363","display_name":"Agniva Chowdhury","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chowdhury, Agniva","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136993701","display_name":"Anshumali Shrivastava","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shrivastava, Anshumali","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137043237","display_name":"Joseph E. Gonzalez","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Gonzalez, Joseph E.","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5137076723","display_name":"Ion Stoica","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Stoica, Ion","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5000412475","display_name":"Kumar Krishna Agrawal","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Agrawal, Kumar Krishna","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5137021734","display_name":"Aditya Desai","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Desai, Aditya","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/T10028","display_name":"Topic Modeling","score":0.21060000360012054,"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/T10028","display_name":"Topic Modeling","score":0.21060000360012054,"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/T11181","display_name":"Advanced Data Storage Technologies","score":0.09849999845027924,"subfield":{"id":"https://openalex.org/subfields/1705","display_name":"Computer Networks and Communications"},"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":0.08699999749660492,"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/inference","display_name":"Inference","score":0.7804999947547913},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.7623000144958496},{"id":"https://openalex.org/keywords/process","display_name":"Process (computing)","score":0.5289999842643738},{"id":"https://openalex.org/keywords/position","display_name":"Position (finance)","score":0.5270000100135803},{"id":"https://openalex.org/keywords/dimension","display_name":"Dimension (graph theory)","score":0.5182999968528748},{"id":"https://openalex.org/keywords/space","display_name":"Space (punctuation)","score":0.39489999413490295},{"id":"https://openalex.org/keywords/illusion","display_name":"Illusion","score":0.38429999351501465},{"id":"https://openalex.org/keywords/empirical-research","display_name":"Empirical research","score":0.35760000348091125}],"concepts":[{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.7804999947547913},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.7623000144958496},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7182999849319458},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5403000116348267},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.5289999842643738},{"id":"https://openalex.org/C198082294","wikidata":"https://www.wikidata.org/wiki/Q3399648","display_name":"Position (finance)","level":2,"score":0.5270000100135803},{"id":"https://openalex.org/C33676613","wikidata":"https://www.wikidata.org/wiki/Q13415176","display_name":"Dimension (graph theory)","level":2,"score":0.5182999968528748},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4562999904155731},{"id":"https://openalex.org/C2778572836","wikidata":"https://www.wikidata.org/wiki/Q380933","display_name":"Space (punctuation)","level":2,"score":0.39489999413490295},{"id":"https://openalex.org/C184047640","wikidata":"https://www.wikidata.org/wiki/Q182593","display_name":"Illusion","level":2,"score":0.38429999351501465},{"id":"https://openalex.org/C120936955","wikidata":"https://www.wikidata.org/wiki/Q2155640","display_name":"Empirical research","level":2,"score":0.35760000348091125},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.3490999937057495},{"id":"https://openalex.org/C76188268","wikidata":"https://www.wikidata.org/wiki/Q1783165","display_name":"Context effect","level":3,"score":0.34139999747276306},{"id":"https://openalex.org/C33566652","wikidata":"https://www.wikidata.org/wiki/Q1065927","display_name":"Theme (computing)","level":2,"score":0.3407000005245209},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.3400999903678894},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.33640000224113464},{"id":"https://openalex.org/C183322885","wikidata":"https://www.wikidata.org/wiki/Q17007702","display_name":"Context model","level":3,"score":0.30809998512268066},{"id":"https://openalex.org/C98184364","wikidata":"https://www.wikidata.org/wiki/Q1780131","display_name":"Argument (complex analysis)","level":2,"score":0.30790001153945923},{"id":"https://openalex.org/C78780964","wikidata":"https://www.wikidata.org/wiki/Q7233193","display_name":"Position paper","level":2,"score":0.3003000020980835},{"id":"https://openalex.org/C166052673","wikidata":"https://www.wikidata.org/wiki/Q83021","display_name":"Empirical evidence","level":2,"score":0.27630001306533813},{"id":"https://openalex.org/C7149132","wikidata":"https://www.wikidata.org/wiki/Q1377840","display_name":"Forgetting","level":2,"score":0.27570000290870667},{"id":"https://openalex.org/C165696696","wikidata":"https://www.wikidata.org/wiki/Q11287","display_name":"Exploit","level":2,"score":0.263700008392334},{"id":"https://openalex.org/C158600405","wikidata":"https://www.wikidata.org/wiki/Q5054566","display_name":"Causal inference","level":2,"score":0.2581999897956848}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.24168","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.24168","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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.24168","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.24168","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":"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"},"sustainable_development_goals":[{"display_name":"Industry, innovation and infrastructure","id":"https://metadata.un.org/sdg/9","score":0.47081848978996277}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Sparsity":[0],"has":[1],"long":[2,98],"been":[3],"a":[4,97,100,108,148,232,236,254],"central":[5],"theme":[6],"in":[7,13,65,96,129],"LLM":[8,19,62,240],"efficiency,":[9],"but":[10,67,234],"its":[11],"role":[12],"context":[14,72,99,139,158,228],"processing":[15,206],"remains":[16],"unresolved.":[17],"As":[18],"workloads":[20],"shift":[21],"toward":[22],"longer":[23],"contexts":[24],"and":[25,30,55,57,83,141,178,243,251,253],"agentic":[26,179],"interactions,":[27],"the":[28,39,59,71,89,117,221],"compute":[29],"memory":[31],"bottlenecks":[32],"of":[33,41,61,81,111,127,169],"attention":[34,93,105],"become":[35],"increasingly":[36],"critical,":[37],"raising":[38],"question":[40],"whether":[42],"these":[43,51,224],"constraints":[44,52],"are":[45,53,160],"fundamental.":[46],"Our":[47],"position":[48,75,226],"is":[49,76,188,248],"that":[50,58,185,247],"artificial":[54],"unnecessary,":[56],"future":[60,258],"inference":[63],"lies":[64],"extreme":[66,227],"principled":[68,237],"sparsity":[69,128,143,166,215,229],"along":[70],"dimension.":[73],"This":[74],"supported":[77],"by":[78,207],"several":[79],"strands":[80],"empirical":[82],"theoretical":[84],"evidence.":[85],"First,":[86],"we":[87,122,182],"find":[88],"insistence":[90],"on":[91,217],"dense":[92],"unreasonable,":[94],"since":[95],"query":[101],"effectively":[102],"projects":[103],"O(N)":[104],"information":[106],"into":[107],"hidden":[109],"space":[110],"dimension":[112],"d":[113],"<<":[114],"N,":[115],"making":[116],"process":[118],"inherently":[119],"lossy.":[120],"Second,":[121],"perform":[123],"an":[124],"extensive":[125],"study":[126],"LLMs":[130],"spanning":[131],"20":[132],"models":[133],"across":[134,167],"five":[135],"model":[136],"families,":[137],"varying":[138,170],"lengths,":[140],"different":[142],"levels.":[144],"We":[145],"empirically":[146],"demonstrate":[147],"strong":[149],"trend:":[150],"current":[151,186],"LLMs,":[152],"despite":[153],"not":[154,230],"being":[155],"trained":[156],"for":[157,239,257],"sparsity,":[159],"remarkably":[161],"robust":[162],"to":[163,191,209],"inference-time":[164],"decode":[165,202],"tasks":[168],"complexity,":[171],"including":[172],"retrieval,":[173],"multi-hop":[174],"QA,":[175],"mathematical":[176],"reasoning,":[177],"coding.":[180],"Importantly,":[181],"also":[183],"show":[184],"hardware":[187,218],"already":[189],"sufficient":[190],"realize":[192],"substantial":[193],"gains":[194],"from":[195],"this":[196],"sparsity.":[197],"For":[198],"example,":[199],"our":[200],"sparse":[201],"kernels":[203],"accelerate":[204],"large-context":[205],"up":[208],"10x":[210],"over":[211],"FlashInfer":[212],"at":[213],"50x":[214],"levels":[216],"such":[219],"as":[220,231,235],"H100.":[222],"Overall,":[223],"results":[225],"heuristic,":[233],"foundation":[238],"inference,":[241],"training,":[242],"architecture":[244],"design:":[245],"one":[246],"both":[249],"feasible":[250],"beneficial,":[252],"compelling":[255],"direction":[256],"systems.":[259]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-27T00:00:00"}