{"id":"https://openalex.org/W7160979246","doi":"https://doi.org/10.48550/arxiv.2605.11605","title":"Keep What Audio Cannot Say: Context-Preserving Token Pruning for Omni-LLMs","display_name":"Keep What Audio Cannot Say: Context-Preserving Token Pruning for Omni-LLMs","publication_year":2026,"publication_date":"2026-05-12","ids":{"openalex":"https://openalex.org/W7160979246","doi":"https://doi.org/10.48550/arxiv.2605.11605"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.11605","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.11605","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.2605.11605","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5062754007","display_name":"Chaeyoung Jung","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jung, Chaeyoung","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5003708446","display_name":"Kyeongha Rho","orcid":"https://orcid.org/0009-0001-6487-5418"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Rho, Kyeongha","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136044425","display_name":"Joon Son Chung","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chung, Joon Son","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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.8245999813079834,"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":0.8245999813079834,"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":0.06129999831318855,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.021800000220537186,"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/security-token","display_name":"Security token","score":0.8575000166893005},{"id":"https://openalex.org/keywords/pruning","display_name":"Pruning","score":0.7998999953269958},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.6334999799728394},{"id":"https://openalex.org/keywords/semantics","display_name":"Semantics (computer science)","score":0.6090999841690063},{"id":"https://openalex.org/keywords/reduction","display_name":"Reduction (mathematics)","score":0.49939998984336853}],"concepts":[{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.8575000166893005},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8410999774932861},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.7998999953269958},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6334999799728394},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.6090999841690063},{"id":"https://openalex.org/C111335779","wikidata":"https://www.wikidata.org/wiki/Q3454686","display_name":"Reduction (mathematics)","level":2,"score":0.49939998984336853},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.4977000057697296},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.48089998960494995},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.44130000472068787},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.3862000107765198},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.3359000086784363},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.3240000009536743},{"id":"https://openalex.org/C183322885","wikidata":"https://www.wikidata.org/wiki/Q17007702","display_name":"Context model","level":3,"score":0.2662999927997589}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.11605","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.11605","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.2605.11605","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.11605","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":{"Omnimodal":[0],"Large":[1],"Language":[2],"Models":[3],"(Omni-LLMs)":[4],"incur":[5],"substantial":[6],"computational":[7],"overhead":[8],"due":[9],"to":[10,130],"the":[11,42],"large":[12],"number":[13],"of":[14,198,204],"multimodal":[15],"input":[16,205],"tokens":[17,37,116,129],"they":[18],"process,":[19],"making":[20],"token":[21,83,99],"reduction":[22,84],"essential":[23],"for":[24,41,65,69],"real-world":[25],"deployment.":[26],"Existing":[27],"Omni-LLM":[28,82],"pruning":[29,100,182,185,202],"methods":[30,183],"typically":[31],"reduce":[32],"this":[33,78,104],"cost":[34],"by":[35],"selecting":[36],"that":[38,57,135],"are":[39,120],"important":[40],"current":[43],"query":[44],"or":[45,68],"strongly":[46],"aligned":[47,73],"with":[48],"cross-modal":[49,92],"cues.":[50,75],"However,":[51],"such":[52],"strategies":[53],"can":[54],"discard":[55],"evidence":[56],"falls":[58],"outside":[59],"these":[60],"criteria,":[61],"even":[62],"when":[63],"needed":[64],"different":[66],"questions":[67],"understanding":[70],"context":[71,89],"beyond":[72],"audio-visual":[74,88,176],"To":[76],"address":[77],"limitation,":[79],"we":[80],"reframe":[81],"as":[85],"preserving":[86],"broad":[87],"while":[90,125,184,201],"removing":[91],"redundancy.":[93],"We":[94],"propose":[95],"ContextGuard,":[96],"an":[97,160],"inference-time":[98,181],"framework":[101,151],"built":[102],"on":[103,189,196],"principle.":[105],"ContextGuard":[106,178,192],"predicts":[107],"coarse":[108,118],"visual":[109,133],"semantics":[110,119],"from":[111,123],"audio":[112,136],"and":[113,157,167,171],"prunes":[114],"video":[115,128,148],"whose":[117],"likely":[121],"recoverable":[122],"audio,":[124],"retaining":[126],"additional":[127],"preserve":[131],"localized":[132],"details":[134],"alone":[137],"cannot":[138],"specify.":[139],"For":[140],"further":[141],"compression,":[142],"our":[143],"method":[144],"merges":[145],"temporally":[146],"similar":[147],"tokens.":[149,187,206],"The":[150],"requires":[152],"no":[153],"downstream":[154],"LLM":[155],"fine-tuning":[156],"uses":[158],"only":[159],"independently":[161],"trained":[162],"lightweight":[163],"predictor.":[164],"On":[165],"Qwen2.5-Omni":[166,190],"Video-SALMONN2+":[168],"at":[169],"3B":[170],"7B":[172],"scales":[173],"across":[174],"six":[175,199],"benchmarks,":[177],"outperforms":[179],"prior":[180],"more":[186],"Notably,":[188],"7B,":[191],"achieves":[193],"full-token-level":[194],"performance":[195],"five":[197],"benchmarks":[200],"55%":[203]},"counts_by_year":[],"updated_date":"2026-07-28T07:46:37.118299","created_date":"2026-05-14T00:00:00"}