{"id":"https://openalex.org/W7204158490","doi":"https://doi.org/10.18653/2026.emnlp-main.1212","title":"Reservoir of Importance: Learning Semi-Structured Sparsity with Differentiable Subset Sampling","display_name":"Reservoir of Importance: Learning Semi-Structured Sparsity with Differentiable Subset Sampling","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7204158490","doi":"https://doi.org/10.18653/2026.emnlp-main.1212"},"language":"en","primary_location":{"id":"doi:10.18653/2026.emnlp-main.1212","doi":"https://doi.org/10.18653/2026.emnlp-main.1212","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.emnlp-main.1212","pdf_url":"https://aclanthology.org/2026.emnlp-main.1212.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","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":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","is_traditional":true,"indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"bronze","oa_url":"https://aclanthology.org/2026.emnlp-main.1212.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5148149843","display_name":"Ha Dinh","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ha Dinh","raw_affiliation_strings":["Viettel AI , Viettel Group , Vietnam"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Viettel AI , Viettel Group , Vietnam","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5118955207","display_name":"Xuan Duy Ta","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xuan Duy Ta","raw_affiliation_strings":["Viettel AI , Viettel Group , Vietnam"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Viettel AI , Viettel Group , Vietnam","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125709665","display_name":"Khoat Than","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4405277486","display_name":"Tr\u01b0\u1eddng \u0110\u1ea1i h\u1ecdc Khoa h\u1ecdc v\u00e0 C\u00f4ng ngh\u1ec7 H\u00e0 N\u1ed9i","ror":"https://ror.org/02wdwrr89","country_code":"VN","type":"education","lineage":["https://openalex.org/I4405277486"]},{"id":"https://openalex.org/I94518387","display_name":"Hanoi University of Science and Technology","ror":"https://ror.org/04nyv3z04","country_code":"VN","type":"education","lineage":["https://openalex.org/I94518387"]}],"countries":["VN"],"is_corresponding":false,"raw_author_name":"Khoat Than","raw_affiliation_strings":["Hanoi University of Science and Technology , Hanoi , Vietnam"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hanoi University of Science and Technology , Hanoi , Vietnam","institution_ids":["https://openalex.org/I4405277486","https://openalex.org/I94518387"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5014247980","display_name":"Khac-Hoai Nam Bui","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Khac-Hoai Nam Bui","raw_affiliation_strings":["Viettel AI , Viettel Group , Vietnam"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Viettel AI , Viettel Group , Vietnam","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.59366045,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"27702","last_page":"27718"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.9771000146865845,"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":0.9771000146865845,"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":0.008700000122189522,"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":0.001500000013038516,"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/qwen2-5","display_name":"Qwen2.5","score":0.8309999704360962}],"concepts":[{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.7901999950408936},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.7117000222206116},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6692000031471252},{"id":"https://openalex.org/C2777735758","wikidata":"https://www.wikidata.org/wiki/Q817765","display_name":"Path (computing)","level":2,"score":0.5892000198364258},{"id":"https://openalex.org/C5274069","wikidata":"https://www.wikidata.org/wiki/Q2285707","display_name":"Categorical variable","level":2,"score":0.5418999791145325},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.5390999913215637},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.49779999256134033},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.4722000062465668},{"id":"https://openalex.org/C202615002","wikidata":"https://www.wikidata.org/wiki/Q783507","display_name":"Differentiable function","level":2,"score":0.46070000529289246},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.421099990606308},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.4205000102519989},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.4203999936580658},{"id":"https://openalex.org/C176217482","wikidata":"https://www.wikidata.org/wiki/Q860554","display_name":"Metric (unit)","level":2,"score":0.36059999465942383},{"id":"https://openalex.org/C2776401178","wikidata":"https://www.wikidata.org/wiki/Q12050496","display_name":"Feature (linguistics)","level":2,"score":0.34209999442100525},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.3203999996185303},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.3181999921798706},{"id":"https://openalex.org/C77618280","wikidata":"https://www.wikidata.org/wiki/Q1155772","display_name":"Scheme (mathematics)","level":2,"score":0.3172000050544739},{"id":"https://openalex.org/C52740198","wikidata":"https://www.wikidata.org/wiki/Q1539564","display_name":"Importance sampling","level":3,"score":0.2948000133037567},{"id":"https://openalex.org/C28719098","wikidata":"https://www.wikidata.org/wiki/Q44946","display_name":"Point (geometry)","level":2,"score":0.2847999930381775},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2838999927043915},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.267300009727478},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.2669999897480011},{"id":"https://openalex.org/C179799912","wikidata":"https://www.wikidata.org/wiki/Q205084","display_name":"Computational complexity theory","level":2,"score":0.2612999975681305},{"id":"https://openalex.org/C2781089630","wikidata":"https://www.wikidata.org/wiki/Q21856745","display_name":"Realization (probability)","level":2,"score":0.25}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.18653/2026.emnlp-main.1212","doi":"https://doi.org/10.18653/2026.emnlp-main.1212","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.emnlp-main.1212","pdf_url":"https://aclanthology.org/2026.emnlp-main.1212.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","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":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2608.23048","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2608.23048","pdf_url":"https://arxiv.org/pdf/2608.23048","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.2608.23048","doi":"https://doi.org/10.48550/arxiv.2608.23048","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2608.23048","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":"doi:10.18653/2026.emnlp-main.1212","doi":"https://doi.org/10.18653/2026.emnlp-main.1212","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.emnlp-main.1212","pdf_url":"https://aclanthology.org/2026.emnlp-main.1212.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","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":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7204158490.pdf","grobid_xml":"https://content.openalex.org/works/W7204158490.grobid-xml"},"referenced_works_count":1,"referenced_works":["https://openalex.org/W4416050145"],"related_works":["https://openalex.org/W4415799245","https://openalex.org/W4407123405","https://openalex.org/W4415114220","https://openalex.org/W7160609909","https://openalex.org/W4396780513","https://openalex.org/W7161683482","https://openalex.org/W7214797770","https://openalex.org/W4415247909","https://openalex.org/W6929426620","https://openalex.org/W7134065531"],"abstract_inverted_index":{"Semi-structured":[0],"N":[1,79,156],":M":[2,80,157],"sparsity":[3,33,63,88,127,158],"has":[4],"emerged":[5],"as":[6],"a":[7,42,56,84,109,161],"practical":[8,162],"direction":[9],"for":[10,87],"accelerating":[11],"large":[12,29],"language":[13],"models":[14,30],"(LLMs).However,":[15],"existing":[16],"learnable-mask":[17],"approaches":[18],"incur":[19],"substantial":[20],"parameter":[21],"and":[22,31,91,116,151],"memory":[23,119,148],"overhead,":[24],"limiting":[25],"their":[26],"scalability":[27,152],"to":[28,96,106,153],"aggressive":[32,155],"regimes.In":[34],"this":[35],"work,":[36],"we":[37],"revisit":[38],"semistructured":[39],"pruning":[40,59],"from":[41,103],"perspective":[43],"that":[44,61,71,141],"reconciles":[45],"efficiency":[46],"with":[47,125,146],"scalability.We":[48],"propose":[49],"Reservoir":[50],"of":[51,133],"Importance":[52],"(RoI)":[53],"1":[54],",":[55],"lightweight":[57],"semi-structured":[58],"framework":[60],"learns":[62],"masks":[64],"through":[65],"differentiable":[66],"subset":[67],"sampling.Unlike":[68],"prior":[69],"methods":[70],"model":[72],"full":[73],"categorical":[74],"distributions":[75],"over":[76],"all":[77],"feasible":[78],"patterns,":[81,159],"RoI":[82,111,142],"introduces":[83],"compact-logit":[85],"parameterization":[86],"mask":[89],"learning":[90],"performs":[92],"sampling":[93],"without":[94],"replacement":[95],"select":[97],"masks,":[98],"thereby":[99],"reducing":[100],"trainable":[101],"parameters":[102,115],"combinatorial":[104],"complexity":[105],"O(M":[107],").As":[108],"result,":[110],"requires":[112],"1.5-8.75\u00d7fewer":[113],"learnable":[114],"significantly":[117],"lower":[118],"cost,":[120],"while":[121],"remaining":[122],"fully":[123],"aligned":[124],"hardware-friendly":[126],"patterns.Extensive":[128],"evaluations":[129],"across":[130],"multiple":[131],"scales":[132],"the":[134],"Qwen2.5":[135],"LLM":[136,166],"family":[137],"(0.5-7B":[138],"parameters)":[139],"demonstrate":[140],"achieves":[143],"competitive":[144],"performance":[145],"strong":[147],"efficiency,":[149],"stability,":[150],"more":[154],"offering":[160],"path":[163],"toward":[164],"efficient":[165],"deployment.":[167]},"counts_by_year":[],"updated_date":"2026-11-08T12:12:28.789561","created_date":"2026-08-26T00:00:00"}