{"id":"https://openalex.org/W7221746331","doi":"https://doi.org/10.48550/arxiv.2610.15394","title":"WUSPruner: Second-Moment-Optimal Transforms for Structured Pruning","display_name":"WUSPruner: Second-Moment-Optimal Transforms for Structured Pruning","publication_year":2026,"publication_date":"2026-10-11","ids":{"openalex":"https://openalex.org/W7221746331","doi":"https://doi.org/10.48550/arxiv.2610.15394"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2610.15394","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.15394","pdf_url":"https://arxiv.org/pdf/2610.15394","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":null,"license_id":null,"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/2610.15394","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5158749175","display_name":"Shengkun Tang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tang, Shengkun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5083822059","display_name":"Dan Alistarh","orcid":"https://orcid.org/0000-0003-3650-940X","observed_orcids":["https://orcid.org/0000-0003-3650-940X"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Alistarh, Dan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5158725980","display_name":"Zhiqiang Shen","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Shen, Zhiqiang","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":true,"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":0.9672999978065491,"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":0.9672999978065491,"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":0.013500000350177288,"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":0.0017000000225380063,"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/structured-pruning","display_name":"structured pruning","score":0.9359999895095825},{"id":"https://openalex.org/keywords/large-language-models","display_name":"large language models","score":0.9129999876022339},{"id":"https://openalex.org/keywords/basis-selection","display_name":"basis selection","score":0.7549999952316284},{"id":"https://openalex.org/keywords/spectral-decomposition","display_name":"spectral decomposition","score":0.5210000276565552},{"id":"https://openalex.org/keywords/model-compression","display_name":"model compression","score":0.5659999847412109},{"id":"https://openalex.org/keywords/inference-latency-reduction","display_name":"inference latency reduction","score":0.671999990940094}],"concepts":[{"id":"https://openalex.org/C108010975","wikidata":"https://www.wikidata.org/wiki/Q500094","display_name":"Pruning","level":2,"score":0.7559999823570251},{"id":"https://openalex.org/C12426560","wikidata":"https://www.wikidata.org/wiki/Q189569","display_name":"Basis (linear algebra)","level":2,"score":0.6087999939918518},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5914999842643738},{"id":"https://openalex.org/C204241405","wikidata":"https://www.wikidata.org/wiki/Q461499","display_name":"Transformation (genetics)","level":3,"score":0.5250999927520752},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5110999941825867},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5073000192642212},{"id":"https://openalex.org/C96442724","wikidata":"https://www.wikidata.org/wiki/Q242188","display_name":"Invertible matrix","level":2,"score":0.4869999885559082},{"id":"https://openalex.org/C48044578","wikidata":"https://www.wikidata.org/wiki/Q727490","display_name":"Scalability","level":2,"score":0.46059998869895935},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4415000081062317},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.43140000104904175},{"id":"https://openalex.org/C92757383","wikidata":"https://www.wikidata.org/wiki/Q382497","display_name":"Affine transformation","level":2,"score":0.3903999924659729},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.35350000858306885},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.30640000104904175},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.30410000681877136},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.28790000081062317},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.2822999954223633},{"id":"https://openalex.org/C125411270","wikidata":"https://www.wikidata.org/wiki/Q18653","display_name":"Encoding (memory)","level":2,"score":0.2662000060081482},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.26350000500679016},{"id":"https://openalex.org/C187834632","wikidata":"https://www.wikidata.org/wiki/Q188804","display_name":"Factorization","level":2,"score":0.25839999318122864},{"id":"https://openalex.org/C5917680","wikidata":"https://www.wikidata.org/wiki/Q2621825","display_name":"Basis function","level":2,"score":0.2554999887943268},{"id":"https://openalex.org/C42355184","wikidata":"https://www.wikidata.org/wiki/Q1361088","display_name":"Matrix decomposition","level":3,"score":0.25040000677108765}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2610.15394","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.15394","pdf_url":"https://arxiv.org/pdf/2610.15394","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2610.15394","doi":"https://doi.org/10.48550/arxiv.2610.15394","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2610.15394","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:2610.15394","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2610.15394","pdf_url":"https://arxiv.org/pdf/2610.15394","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7221746331.pdf","grobid_xml":"https://content.openalex.org/works/W7221746331.grobid-xml"},"referenced_works_count":17,"referenced_works":["https://openalex.org/W2794325560","https://openalex.org/W3168867926","https://openalex.org/W3195140297","https://openalex.org/W3210904070","https://openalex.org/W4380715573","https://openalex.org/W4389650671","https://openalex.org/W4391988255","https://openalex.org/W4392575839","https://openalex.org/W4392780917","https://openalex.org/W4393147854","https://openalex.org/W4401306886","https://openalex.org/W4404987838","https://openalex.org/W4412505604","https://openalex.org/W4414939355","https://openalex.org/W4416956316","https://openalex.org/W7133245307","https://openalex.org/W7160893354"],"related_works":["https://openalex.org/W3172288629","https://openalex.org/W4402683876","https://openalex.org/W4415247909","https://openalex.org/W7171568314","https://openalex.org/W4406975730","https://openalex.org/W4386395712","https://openalex.org/W4402671801","https://openalex.org/W7155451756","https://openalex.org/W2979691890","https://openalex.org/W4402670450"],"abstract_inverted_index":{"Structured":[0],"pruning":[1,31,58,162,189],"is":[2,32],"typically":[3],"formulated":[4],"as":[5,67],"identifying":[6],"which":[7,30],"model":[8],"dimensions":[9],"to":[10,140,181],"remove,":[11],"often":[12],"within":[13],"the":[14,19,27,42,76,84,111,122,125,137],"model's":[15],"original":[16],"representation.":[17],"However,":[18],"effect":[20],"of":[21,59,119],"coordinate":[22,48,152],"deletion":[23,138],"fundamentally":[24],"depends":[25],"on":[26,165,174],"basis":[28,43,65],"in":[29,87,207],"performed.":[33],"This":[34],"raises":[35],"a":[36,53,68,91,148,160],"more":[37],"fundamental":[38],"question:":[39],"how":[40],"should":[41],"be":[44],"chosen":[45],"specifically":[46],"for":[47,56,151],"deletion?":[49],"We":[50,63,101],"introduce":[51],"WUSPruner,":[52],"transform-based":[54],"framework":[55],"structured":[57,188],"large":[60],"language":[61,192],"models.":[62],"formulate":[64],"selection":[66],"local":[69,112],"linear":[70],"reconstruction":[71,77,113],"problem":[72],"that":[73,103],"directly":[74],"minimizes":[75,110],"loss":[78,114],"induced":[79],"by":[80,83,155],"pruning.":[81],"Inspired":[82],"second-moment":[85],"transformation":[86],"WUSH,":[88],"we":[89,158],"consider":[90],"WUS":[92,108,167],"transform":[93,109],"constructed":[94],"jointly":[95],"from":[96,179],"activation":[97,126],"and":[98,127,134,146,169,176,198,210],"weight":[99,128],"statistics.":[100],"show":[102,184],"under":[104],"positive-definite":[105],"second":[106,129],"moments,":[107],"over":[115,187],"all":[116],"invertible":[117],"changes":[118],"basis.":[120],"In":[121],"resulting":[123],"basis,":[124],"moments":[130],"are":[131],"simultaneously":[132],"diagonalized":[133],"balanced,":[135],"causing":[136],"error":[139],"decompose":[141],"into":[142],"independent":[143],"spectral":[144],"contributions":[145],"yielding":[147],"principled":[149],"criterion":[150],"selection.":[153],"Guided":[154],"this":[156],"result,":[157],"develop":[159],"scalable":[161],"method":[163],"based":[164],"blockwise":[166],"transformations":[168],"compact":[170],"input":[171],"projections.":[172],"Experiments":[173],"Llama":[175],"Qwen":[177],"models":[178],"8B":[180],"72B":[182],"parameters":[183],"consistent":[185],"improvements":[186],"baselines":[190],"across":[191],"modeling,":[193],"knowledge,":[194],"commonsense":[195],"reasoning,":[196],"mathematics,":[197],"code":[199],"generation":[200],"benchmarks,":[201],"while":[202],"also":[203],"delivering":[204],"practical":[205],"reductions":[206],"inference":[208],"latency":[209],"memory":[211],"usage.":[212]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-10-15T00:00:00"}