{"id":"https://openalex.org/W7128689792","doi":"https://doi.org/10.48448/5019-9964","title":"CasMoE: A Cascaded Framework for Efficient MoE Inference on Resource-constrained Devices","display_name":"CasMoE: A Cascaded Framework for Efficient MoE Inference on Resource-constrained Devices","publication_year":2026,"publication_date":"2026-01-07","ids":{"openalex":"https://openalex.org/W7128689792","doi":"https://doi.org/10.48448/5019-9964"},"language":"en","primary_location":{"id":"doi:10.48448/5019-9964","doi":"https://doi.org/10.48448/5019-9964","is_oa":true,"landing_page_url":"https://doi.org/10.48448/5019-9964","pdf_url":null,"source":{"id":"https://openalex.org/S7407053148","display_name":"Underline Science Inc.","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"Audiovisual"},"volume":null,"type":"other","is_traditional":false,"indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48448/5019-9964","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5125741655","display_name":"Association for Artificial Intelligence 2026","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Association for Artificial Intelligence 2026","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125705092","display_name":"Xiaoheng Deng","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Deng, Xiaoheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125718541","display_name":"Lixin Duan","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Duan, Lixin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125734767","display_name":"Haowen He","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"He, Haowen","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125696741","display_name":"SHAOHUA WAN","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"WAN, SHAOHUA","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125764347","display_name":"Chengcheng Wang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Chengcheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5125691757","display_name":"Liang Zhao","orcid":"https://orcid.org/0000-0002-1502-6604","observed_orcids":["https://orcid.org/0000-0002-1502-6604"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhao, Liang","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":0.8769999742507935,"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":0.8769999742507935,"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":0.06040000170469284,"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":0.02889999933540821,"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/mixture-of-experts","display_name":"mixture of experts","score":0.9660000205039978},{"id":"https://openalex.org/keywords/resource-constrained-devices","display_name":"resource-constrained devices","score":0.9629999995231628},{"id":"https://openalex.org/keywords/gating-mechanism","display_name":"gating mechanism","score":0.5270000100135803},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.7699999809265137},{"id":"https://openalex.org/keywords/memory-efficiency","display_name":"memory efficiency","score":0.4350000023841858}],"concepts":[{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.76910001039505},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7555999755859375},{"id":"https://openalex.org/C58328972","wikidata":"https://www.wikidata.org/wiki/Q184609","display_name":"Expert system","level":2,"score":0.5849000215530396},{"id":"https://openalex.org/C2779960059","wikidata":"https://www.wikidata.org/wiki/Q7113681","display_name":"Overhead (engineering)","level":2,"score":0.5792999863624573},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.5497999787330627},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5105000138282776},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4934999942779541},{"id":"https://openalex.org/C165464430","wikidata":"https://www.wikidata.org/wiki/Q1570441","display_name":"Parameterized complexity","level":2,"score":0.4887000024318695},{"id":"https://openalex.org/C46743427","wikidata":"https://www.wikidata.org/wiki/Q1341685","display_name":"Inference engine","level":3,"score":0.4702000021934509},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.38499999046325684},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.37700000405311584},{"id":"https://openalex.org/C105002631","wikidata":"https://www.wikidata.org/wiki/Q4833645","display_name":"Subject-matter expert","level":3,"score":0.3596999943256378},{"id":"https://openalex.org/C21200559","wikidata":"https://www.wikidata.org/wiki/Q7451068","display_name":"Sensitivity (control systems)","level":2,"score":0.35910001397132874},{"id":"https://openalex.org/C2777472644","wikidata":"https://www.wikidata.org/wiki/Q16968992","display_name":"Approximate inference","level":3,"score":0.3391000032424927},{"id":"https://openalex.org/C2777851325","wikidata":"https://www.wikidata.org/wiki/Q7094102","display_name":"Online model","level":2,"score":0.31150001287460327},{"id":"https://openalex.org/C2777904410","wikidata":"https://www.wikidata.org/wiki/Q7397","display_name":"Software","level":2,"score":0.30070000886917114},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.298799991607666},{"id":"https://openalex.org/C86610423","wikidata":"https://www.wikidata.org/wiki/Q1925081","display_name":"Metamodeling","level":2,"score":0.25780001282691956},{"id":"https://openalex.org/C155846161","wikidata":"https://www.wikidata.org/wiki/Q1143367","display_name":"Graphical model","level":2,"score":0.25189998745918274}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48448/5019-9964","doi":"https://doi.org/10.48448/5019-9964","is_oa":true,"landing_page_url":"https://doi.org/10.48448/5019-9964","pdf_url":null,"source":{"id":"https://openalex.org/S7407053148","display_name":"Underline Science Inc.","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"Audiovisual"}],"best_oa_location":{"id":"doi:10.48448/5019-9964","doi":"https://doi.org/10.48448/5019-9964","is_oa":true,"landing_page_url":"https://doi.org/10.48448/5019-9964","pdf_url":null,"source":{"id":"https://openalex.org/S7407053148","display_name":"Underline Science Inc.","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"Audiovisual"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4403006781","https://openalex.org/W7137967374","https://openalex.org/W6929720056","https://openalex.org/W7154312709","https://openalex.org/W4417250987","https://openalex.org/W7082260013","https://openalex.org/W4404354159","https://openalex.org/W7206202243","https://openalex.org/W7077974163","https://openalex.org/W7166741390"],"abstract_inverted_index":{"The":[0],"Mixture-of-Experts":[1],"(MoE)":[2],"architecture":[3],"has":[4],"emerged":[5],"as":[6,50],"a":[7,74,89,102,125,143,161,184],"key":[8],"enabler":[9],"for":[10,80,156],"scaling":[11],"large":[12],"language":[13],"models":[14],"(LLMs),":[15],"empowering":[16],"increased":[17,66],"model":[18,51],"capacity":[19],"with":[20,138],"minimal":[21],"computational":[22],"overhead":[23],"through":[24],"gating-based":[25],"dynamic":[26],"expert":[27,37,96,115,133],"activation.":[28],"However,":[29],"due":[30],"to":[31,62,93,110,141,165,173],"the":[32,99,113,118,122,139,150,176,179],"memory":[33],"demands":[34],"introduced":[35,109],"by":[36],"modules,":[38],"MoE":[39,82,151,158],"inference":[40,83,188,203],"on":[41,84,194],"resource-constrained":[42,85],"devices":[43],"is":[44,108,135,170],"still":[45],"challenging.":[46],"Existing":[47],"methods":[48],"such":[49],"compression":[52],"and":[53,76,181,190],"parameter":[54],"offloading":[55],"provide":[56],"partial":[57],"alleviation":[58],"but":[59],"often":[60],"lead":[61],"reduced":[63],"accuracy":[64],"or":[65],"latency.":[67],"In":[68,98,121],"this":[69],"paper,":[70],"we":[71],"propose":[72],"CasMoE,":[73],"general":[75],"efficient":[77,95],"cascaded":[78],"framework":[79],"accelerating":[81,202],"devices.":[86],"CasMoE":[87],"employs":[88],"two-stage":[90],"offline-online":[91],"approach":[92],"facilitate":[94],"prefetching.":[97],"offline":[100],"stage,":[101,124],"parameterized":[103],"Expert":[104,127],"Activation":[105,128],"Predictor":[106],"(EAP)":[107],"accurately":[111],"predict":[112],"corresponding":[114],"activation":[116],"from":[117],"incoming":[119],"prompt.":[120],"online":[123],"non-parametric":[126],"Matcher":[129],"(EAM)":[130],"supporting":[131],"fast":[132],"retrieval":[134],"then":[136],"integrated":[137],"EAP":[140],"form":[142],"cascade":[144],"planner":[145],"that":[146],"operates":[147],"independently":[148],"of":[149,178],"architecture,":[152],"predicting":[153],"activated":[154],"experts":[155],"all":[157],"layers":[159],"in":[160,201],"single":[162],"pass":[163],"prior":[164],"decoding.":[166],"A":[167],"gating":[168],"mechanism":[169],"also":[171],"incorporated":[172],"dynamically":[174],"adjust":[175],"sensitivity":[177],"EAM":[180],"EAP,":[182],"enabling":[183],"flexible":[185],"trade-off":[186],"between":[187],"efficiency":[189],"quality.":[191],"Extensive":[192],"experiments":[193],"diverse":[195],"downstream":[196],"tasks":[197],"demonstrate":[198],"CasMoE\u2019s":[199],"effectiveness":[200],"while":[204],"preserving":[205],"high":[206],"accuracy.":[207]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-02-13T00:00:00"}