{"id":"https://openalex.org/W7127580109","doi":"https://doi.org/10.48550/arxiv.2602.03051","title":"SAES-SVD: Self-Adaptive Suppression of Accumulated and Local Errors for SVD-based LLM Compression","display_name":"SAES-SVD: Self-Adaptive Suppression of Accumulated and Local Errors for SVD-based LLM Compression","publication_year":2026,"publication_date":"2026-02-03","ids":{"openalex":"https://openalex.org/W7127580109","doi":"https://doi.org/10.48550/arxiv.2602.03051"},"language":null,"primary_location":{"id":"pmh:doi:10.48550/arxiv.2602.03051","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"Article"},"type":"article","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":null,"any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5125041296","display_name":"Xing Hu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hu, Xing","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125038044","display_name":"Dawei Yang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang, Dawei","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125094428","display_name":"Yuan Cheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cheng, Yuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125069908","display_name":"Zhixuan Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Zhixuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5125094306","display_name":"Zukang Xu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xu, Zukang","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":0.0,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":{"value":0.14066077,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"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/T10181","display_name":"Natural Language Processing Techniques","score":0.10729999840259552,"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"}},"topics":[{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.10729999840259552,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.1071000024676323,"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/T10028","display_name":"Topic Modeling","score":0.07100000232458115,"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/compression","display_name":"Compression (physics)","score":0.7008000016212463},{"id":"https://openalex.org/keywords/data-compression","display_name":"Data compression","score":0.5414999723434448},{"id":"https://openalex.org/keywords/weighting","display_name":"Weighting","score":0.5396000146865845},{"id":"https://openalex.org/keywords/data-compression-ratio","display_name":"Data compression ratio","score":0.5117999911308289},{"id":"https://openalex.org/keywords/compression-ratio","display_name":"Compression ratio","score":0.4560999870300293},{"id":"https://openalex.org/keywords/rank","display_name":"Rank (graph theory)","score":0.41449999809265137},{"id":"https://openalex.org/keywords/scale","display_name":"Scale (ratio)","score":0.36239999532699585}],"concepts":[{"id":"https://openalex.org/C180016635","wikidata":"https://www.wikidata.org/wiki/Q2712821","display_name":"Compression (physics)","level":2,"score":0.7008000016212463},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.6295999884605408},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5809000134468079},{"id":"https://openalex.org/C78548338","wikidata":"https://www.wikidata.org/wiki/Q2493","display_name":"Data compression","level":2,"score":0.5414999723434448},{"id":"https://openalex.org/C183115368","wikidata":"https://www.wikidata.org/wiki/Q856577","display_name":"Weighting","level":2,"score":0.5396000146865845},{"id":"https://openalex.org/C94835093","wikidata":"https://www.wikidata.org/wiki/Q3113333","display_name":"Data compression ratio","level":5,"score":0.5117999911308289},{"id":"https://openalex.org/C25797200","wikidata":"https://www.wikidata.org/wiki/Q828137","display_name":"Compression ratio","level":3,"score":0.4560999870300293},{"id":"https://openalex.org/C164226766","wikidata":"https://www.wikidata.org/wiki/Q7293202","display_name":"Rank (graph theory)","level":2,"score":0.41449999809265137},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.36239999532699585},{"id":"https://openalex.org/C126255220","wikidata":"https://www.wikidata.org/wiki/Q141495","display_name":"Mathematical optimization","level":1,"score":0.3422999978065491},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3240000009536743},{"id":"https://openalex.org/C191795146","wikidata":"https://www.wikidata.org/wiki/Q3878446","display_name":"Norm (philosophy)","level":2,"score":0.3154999911785126},{"id":"https://openalex.org/C2779227376","wikidata":"https://www.wikidata.org/wiki/Q6505497","display_name":"Layer (electronics)","level":2,"score":0.3125},{"id":"https://openalex.org/C103088060","wikidata":"https://www.wikidata.org/wiki/Q1062839","display_name":"Error detection and correction","level":2,"score":0.26989999413490295},{"id":"https://openalex.org/C13481523","wikidata":"https://www.wikidata.org/wiki/Q412438","display_name":"Image compression","level":4,"score":0.2671000063419342},{"id":"https://openalex.org/C92207270","wikidata":"https://www.wikidata.org/wiki/Q939253","display_name":"Matrix norm","level":3,"score":0.2547000050544739},{"id":"https://openalex.org/C124851039","wikidata":"https://www.wikidata.org/wiki/Q2665459","display_name":"Compressed sensing","level":2,"score":0.2524000108242035},{"id":"https://openalex.org/C122383733","wikidata":"https://www.wikidata.org/wiki/Q865920","display_name":"Approximation error","level":2,"score":0.2508000135421753}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:doi:10.48550/arxiv.2602.03051","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"Article"},{"id":"doi:10.48550/arxiv.2602.03051","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2602.03051","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:doi:10.48550/arxiv.2602.03051","is_oa":true,"landing_page_url":null,"pdf_url":null,"source":{"id":"https://openalex.org/S4406922384","display_name":"Open MIND","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"Article"},"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":{"The":[0],"rapid":[1],"growth":[2],"in":[3,172],"the":[4,51,58,67,108,160,165,169,175,181,184,188,195,205],"parameter":[5],"scale":[6],"of":[7,96,114,168,187,194],"large":[8],"language":[9],"models":[10],"(LLMs)":[11],"has":[12,30],"created":[13],"a":[14,22,48,80,112,127,199],"high":[15],"demand":[16],"for":[17,148],"efficient":[18],"compression":[19,29,82,109,170,196],"techniques.":[20],"As":[21],"hardware-agnostic":[23],"and":[24,55,89,117,192,217],"highly":[25],"compatible":[26],"technique,":[27],"low-rank":[28,129,166],"been":[31],"widely":[32],"adopted.":[33],"However,":[34],"existing":[35],"methods":[36],"typically":[37],"compress":[38],"each":[39,139],"layer":[40],"independently":[41],"by":[42],"minimizing":[43],"per-layer":[44],"reconstruction":[45,52,88,116],"error,":[46],"overlooking":[47],"critical":[49],"limitation:":[50],"error":[53,91,120],"propagates":[54],"accumulates":[56],"through":[57],"network,":[59],"which":[60,106,136,157],"leads":[61],"to":[62,146,163,179],"amplified":[63],"global":[64],"deviations":[65],"from":[66],"full-precision":[68,144],"baseline.":[69],"To":[70],"address":[71],"this,":[72],"we":[73,125],"propose":[74],"Self-Adaptive":[75],"Error":[76,154],"Suppression":[77,155],"SVD":[78],"(SAES-SVD),":[79],"LLMs":[81],"framework":[83],"that":[84,193,204],"jointly":[85],"optimizes":[86],"intra-layer":[87],"inter-layer":[90],"compensation.":[92,121],"SAES-SVD":[93,226],"is":[94,177,208],"composed":[95],"two":[97],"novel":[98],"components:":[99],"(1)":[100],"Cumulative":[101],"Error-Aware":[102],"Layer":[103],"Compression":[104],"(CEALC),":[105],"formulates":[107],"objective":[110,171,197],"as":[111],"combination":[113],"local":[115],"weighted":[118],"cumulative":[119],"Based":[122],"on":[123,132],"it,":[124],"derive":[126],"closed-form":[128],"solution":[130],"relied":[131],"second-order":[133],"activation":[134],"statistics,":[135],"explicitly":[137],"aligns":[138],"layer's":[140,190],"output":[141,191],"with":[142],"its":[143],"counterpart":[145],"compensate":[147],"accumulated":[149],"errors.":[150],"(2)":[151],"Adaptive":[152],"Collaborative":[153],"(ACES),":[156],"automatically":[158],"adjusts":[159],"weighting":[161],"coefficient":[162,176],"enhance":[164],"structure":[167],"CEALC.":[173],"Specifically,":[174],"optimized":[178],"maximize":[180],"ratio":[182],"between":[183],"Frobenius":[185],"norm":[186],"compressed":[189],"under":[198],"fixed":[200],"rank,":[201],"thus":[202],"ensuring":[203],"rank":[206],"budget":[207],"utilized":[209],"effectively.":[210],"Extensive":[211],"experiments":[212],"across":[213],"multiple":[214],"LLM":[215],"architectures":[216],"tasks":[218],"show":[219],"that,":[220],"without":[221],"fine-tuning":[222],"or":[223],"mixed-rank":[224],"strategies,":[225],"consistently":[227],"improves":[228],"post-compression":[229],"performance.":[230]},"counts_by_year":[],"updated_date":"2026-07-15T18:14:33.161393","created_date":"2026-02-06T00:00:00"}