{"id":"https://openalex.org/W7214747416","doi":"https://doi.org/10.48550/arxiv.2609.31631","title":"OMP-MoE: Efficient Expert Pruning for Mixture-of-Experts LLMs via Orthogonal Matching Pursuit","display_name":"OMP-MoE: Efficient Expert Pruning for Mixture-of-Experts LLMs via Orthogonal Matching Pursuit","publication_year":2026,"publication_date":"2026-09-09","ids":{"openalex":"https://openalex.org/W7214747416","doi":"https://doi.org/10.48550/arxiv.2609.31631"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.31631","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.31631","pdf_url":"https://arxiv.org/pdf/2609.31631","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.31631","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5153455539","display_name":"Dezhi Li","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Li, Dezhi","raw_affiliation_strings":["The Hong Kong University of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology","institution_ids":["https://openalex.org/I200769079"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101983249","display_name":"Lujun Li","orcid":"https://orcid.org/0000-0002-4329-2707","observed_orcids":["https://orcid.org/0000-0002-4329-2707"]},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Li, Lujun","raw_affiliation_strings":["The Hong Kong University of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology","institution_ids":["https://openalex.org/I200769079"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153440099","display_name":"Qiyuan Zhu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Zhu, Qiyuan","raw_affiliation_strings":["The Hong Kong University of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology","institution_ids":["https://openalex.org/I200769079"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153480318","display_name":"Hao Gu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Gu, Hao","raw_affiliation_strings":["The Hong Kong University of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology","institution_ids":["https://openalex.org/I200769079"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5153455553","display_name":"Bei Liu","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Liu, Bei","raw_affiliation_strings":["The Hong Kong University of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology","institution_ids":["https://openalex.org/I200769079"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5021255269","display_name":"Sirui Han","orcid":"https://orcid.org/0000-0001-7303-0671","observed_orcids":["https://orcid.org/0000-0001-7303-0671"]},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Han, Sirui","raw_affiliation_strings":["The Hong Kong University of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology","institution_ids":["https://openalex.org/I200769079"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5045081171","display_name":"Yike Guo","orcid":"https://orcid.org/0000-0002-3075-2161","observed_orcids":["https://orcid.org/0000-0002-3075-2161","https://orcid.org/0009-0005-8401-282X","https://orcid.org/0000-0002-4026-2983"]},"institutions":[{"id":"https://openalex.org/I200769079","display_name":"Hong Kong University of Science and Technology","ror":"https://ror.org/00q4vv597","country_code":"HK","type":"education","lineage":["https://openalex.org/I200769079"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Guo, Yike","raw_affiliation_strings":["The Hong Kong University of Science and Technology"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology","institution_ids":["https://openalex.org/I200769079"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I200769079"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"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":0.98580002784729,"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":0.98580002784729,"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":0.00430000014603138,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.0010000000474974513,"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":0.9750000238418579},{"id":"https://openalex.org/keywords/expert-pruning","display_name":"expert pruning","score":0.7900000214576721},{"id":"https://openalex.org/keywords/orthogonal-matching-pursuit","display_name":"orthogonal matching pursuit","score":0.9229999780654907},{"id":"https://openalex.org/keywords/large-language-model-compression","display_name":"large language model compression","score":0.7450000047683716},{"id":"https://openalex.org/keywords/training-free-pruning","display_name":"training-free pruning","score":0.8399999737739563},{"id":"https://openalex.org/keywords/sparse-signal-recovery","display_name":"sparse signal recovery","score":0.9549999833106995},{"id":"https://openalex.org/keywords/energy-prediction","display_name":"energy prediction","score":0.7080000042915344},{"id":"https://openalex.org/keywords/memory-efficiency","display_name":"memory efficiency","score":0.460999995470047}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7649000287055969},{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.6920999884605408},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.5981000065803528},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5450999736785889},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5367000102996826},{"id":"https://openalex.org/C152124472","wikidata":"https://www.wikidata.org/wiki/Q1204361","display_name":"Redundancy (engineering)","level":2,"score":0.49729999899864197},{"id":"https://openalex.org/C156872377","wikidata":"https://www.wikidata.org/wiki/Q6786281","display_name":"Matching pursuit","level":3,"score":0.46709999442100525},{"id":"https://openalex.org/C2780451532","wikidata":"https://www.wikidata.org/wiki/Q759676","display_name":"Task (project management)","level":2,"score":0.41130000352859497},{"id":"https://openalex.org/C58328972","wikidata":"https://www.wikidata.org/wiki/Q184609","display_name":"Expert system","level":2,"score":0.37790000438690186},{"id":"https://openalex.org/C124101348","wikidata":"https://www.wikidata.org/wiki/Q172491","display_name":"Data mining","level":1,"score":0.3635999858379364},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.36149999499320984},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.33320000767707825},{"id":"https://openalex.org/C22367795","wikidata":"https://www.wikidata.org/wiki/Q7625208","display_name":"Structured prediction","level":2,"score":0.3140999972820282},{"id":"https://openalex.org/C185874996","wikidata":"https://www.wikidata.org/wiki/Q269699","display_name":"Interdependence","level":2,"score":0.31220000982284546},{"id":"https://openalex.org/C2776650193","wikidata":"https://www.wikidata.org/wiki/Q264661","display_name":"Obstacle","level":2,"score":0.299699991941452},{"id":"https://openalex.org/C2779304628","wikidata":"https://www.wikidata.org/wiki/Q3503480","display_name":"Face (sociological concept)","level":2,"score":0.2904999852180481},{"id":"https://openalex.org/C68339613","wikidata":"https://www.wikidata.org/wiki/Q1549489","display_name":"Speedup","level":2,"score":0.2687999904155731},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.26159998774528503},{"id":"https://openalex.org/C52421305","wikidata":"https://www.wikidata.org/wiki/Q1151499","display_name":"Particle filter","level":3,"score":0.25189998745918274}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2609.31631","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.31631","pdf_url":"https://arxiv.org/pdf/2609.31631","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2609.31631","doi":"https://doi.org/10.48550/arxiv.2609.31631","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2609.31631","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,"listed_in":[],"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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.31631","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.31631","pdf_url":"https://arxiv.org/pdf/2609.31631","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,"listed_in":[],"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":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7214747416.pdf","grobid_xml":"https://content.openalex.org/works/W7214747416.grobid-xml"},"referenced_works_count":20,"referenced_works":["https://openalex.org/W2525332836","https://openalex.org/W2581624817","https://openalex.org/W2794325560","https://openalex.org/W3177813494","https://openalex.org/W3210904070","https://openalex.org/W4306808908","https://openalex.org/W4327810158","https://openalex.org/W4390723197","https://openalex.org/W4396815229","https://openalex.org/W4401306886","https://openalex.org/W4405903187","https://openalex.org/W4414742703","https://openalex.org/W4414939355","https://openalex.org/W4416512644","https://openalex.org/W4416875788","https://openalex.org/W6929316192","https://openalex.org/W7075994024","https://openalex.org/W7083686523","https://openalex.org/W7167587142","https://openalex.org/W7197032550"],"related_works":["https://openalex.org/W4404792840","https://openalex.org/W4412888953","https://openalex.org/W7166845002","https://openalex.org/W6910811670","https://openalex.org/W7137282234","https://openalex.org/W4403344761","https://openalex.org/W4403577281","https://openalex.org/W4416875788","https://openalex.org/W7221812033","https://openalex.org/W4416512644"],"abstract_inverted_index":{"Mixture-of-Experts":[0],"(MoE)":[1],"models":[2,9],"enable":[3],"efficient":[4],"scaling":[5],"of":[6,57,163],"large":[7],"language":[8],"but":[10],"face":[11],"critical":[12],"deployment":[13],"challenges":[14],"due":[15],"to":[16],"massive":[17],"memory":[18],"requirements.":[19],"Existing":[20],"pruning":[21,64,153],"methods":[22,150],"either":[23],"incur":[24],"prohibitive":[25],"search":[26,169],"costs":[27],"or":[28],"neglect":[29],"the":[30,63],"dynamic":[31],"interdependencies":[32],"between":[33],"experts.":[34],"To":[35],"address":[36],"these":[37],"challenges,":[38],"we":[39,61,101,120,160],"present":[40],"OMP-MoE,":[41],"a":[42,67,107],"novel":[43],"training-free":[44],"compression":[45],"framework":[46],"for":[47,112],"reducing":[48],"expert":[49,58,83,104,130],"redundancy":[50],"in":[51],"MoE-based":[52],"LLMs.":[53],"Based":[54],"on":[55,133,138],"observations":[56],"contribution":[59],"patterns,":[60],"reformulate":[62],"problem":[65],"as":[66,85],"sparse":[68],"signal":[69],"reconstruction":[70,94,114],"task":[71],"solved":[72],"through":[73,106],"Orthogonal":[74],"Matching":[75],"Pursuit.":[76],"Specifically,":[77],"our":[78],"method":[79],"first":[80],"treats":[81],"individual":[82],"contributions":[84],"dictionary":[86],"atoms":[87],"and":[88,116,142,170],"selects":[89],"experts":[90],"that":[91,110,127],"greedily":[92],"minimize":[93],"error":[95],"with":[96],"linear":[97],"computational":[98],"complexity.":[99],"Then,":[100],"optimize":[102],"cross-layer":[103],"allocation":[105],"water-filling":[108],"strategy":[109],"accounts":[111],"both":[113],"quality":[115],"routing":[117],"stability.":[118],"Finally,":[119],"introduce":[121],"OMP-MoE\u2020,":[122],"an":[123],"adaptive":[124],"inference":[125,172],"mechanism":[126],"dynamically":[128],"adjusts":[129],"activation":[131],"based":[132],"energy":[134],"prediction.":[135],"Comprehensive":[136],"experiments":[137],"Qwen,":[139],"DeepSeek-V2,":[140],"GPT-OSS,":[141],"Mixtral":[143],"MoE":[144],"demonstrate":[145],"consistent":[146],"improvements":[147],"over":[148],"existing":[149],"at":[151,157],"25-50%":[152],"ratios.":[154],"For":[155],"Qwen3-30B-A3B":[156],"50%":[158],"compression,":[159],"retain":[161],"93.3%":[162],"original":[164],"performance,":[165],"achieving":[166],"33$\\times$":[167],"faster":[168],"1.55$\\times$":[171],"speedup.":[173],"Codes":[174],"will":[175],"be":[176],"available":[177],"after":[178],"acceptance.":[179]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-11-10T11:39:43.021295","created_date":"2026-09-30T00:00:00"}