{"id":"https://openalex.org/W7138096296","doi":"https://doi.org/10.1609/aaai.v40i34.40112","title":"SpecQuant: Spectral Decomposition and Adaptive Truncation for Ultra-Low-Bit LLMs Quantization","display_name":"SpecQuant: Spectral Decomposition and Adaptive Truncation for Ultra-Low-Bit LLMs Quantization","publication_year":2026,"publication_date":"2026-03-14","ids":{"openalex":"https://openalex.org/W7138096296","doi":"https://doi.org/10.1609/aaai.v40i34.40112"},"language":null,"primary_location":{"id":"doi:10.1609/aaai.v40i34.40112","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i34.40112","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://doi.org/10.1609/aaai.v40i34.40112","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5125940974","display_name":"Zhixiong Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhixiong Zhao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5017670541","display_name":"Fangxin Liu","orcid":"https://orcid.org/0000-0002-8769-293X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Fangxin Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129658115","display_name":"Junjie Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Junjie Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5081589431","display_name":"Chenyang Guan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chenyang Guan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5107269495","display_name":"Zongwu Wang","orcid":"https://orcid.org/0009-0003-2157-4927"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zongwu Wang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129702826","display_name":"Li Jiang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li Jiang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5129716814","display_name":"Haibing Guan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Haibing Guan","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":"40","issue":"34","first_page":"28786","last_page":"28794"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10036","display_name":"Advanced Neural Network Applications","score":0.24740000069141388,"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":0.24740000069141388,"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/T10201","display_name":"Speech Recognition and Synthesis","score":0.10999999940395355,"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":0.10830000042915344,"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/quantization","display_name":"Quantization (signal processing)","score":0.8364999890327454},{"id":"https://openalex.org/keywords/truncation","display_name":"Truncation (statistics)","score":0.5992000102996826},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.5788000226020813},{"id":"https://openalex.org/keywords/outlier","display_name":"Outlier","score":0.5679000020027161},{"id":"https://openalex.org/keywords/truncation-error","display_name":"Truncation error","score":0.5029000043869019},{"id":"https://openalex.org/keywords/fourier-transform","display_name":"Fourier transform","score":0.413100004196167},{"id":"https://openalex.org/keywords/frequency-domain","display_name":"Frequency domain","score":0.4075999855995178}],"concepts":[{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.8364999890327454},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.6721000075340271},{"id":"https://openalex.org/C106195933","wikidata":"https://www.wikidata.org/wiki/Q7847935","display_name":"Truncation (statistics)","level":2,"score":0.5992000102996826},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.5788000226020813},{"id":"https://openalex.org/C79337645","wikidata":"https://www.wikidata.org/wiki/Q779824","display_name":"Outlier","level":2,"score":0.5679000020027161},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5256999731063843},{"id":"https://openalex.org/C104942944","wikidata":"https://www.wikidata.org/wiki/Q3434686","display_name":"Truncation error","level":2,"score":0.5029000043869019},{"id":"https://openalex.org/C102519508","wikidata":"https://www.wikidata.org/wiki/Q6520159","display_name":"Fourier transform","level":2,"score":0.413100004196167},{"id":"https://openalex.org/C19118579","wikidata":"https://www.wikidata.org/wiki/Q786423","display_name":"Frequency domain","level":2,"score":0.4075999855995178},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.3880000114440918},{"id":"https://openalex.org/C207864730","wikidata":"https://www.wikidata.org/wiki/Q179467","display_name":"Fourier series","level":2,"score":0.35850000381469727},{"id":"https://openalex.org/C2221639","wikidata":"https://www.wikidata.org/wiki/Q2877","display_name":"Discrete cosine transform","level":3,"score":0.33410000801086426},{"id":"https://openalex.org/C103824480","wikidata":"https://www.wikidata.org/wiki/Q185889","display_name":"Time domain","level":2,"score":0.29490000009536743},{"id":"https://openalex.org/C104267543","wikidata":"https://www.wikidata.org/wiki/Q208163","display_name":"Signal processing","level":3,"score":0.29010000824928284},{"id":"https://openalex.org/C186370098","wikidata":"https://www.wikidata.org/wiki/Q442787","display_name":"Energy (signal processing)","level":2,"score":0.28700000047683716},{"id":"https://openalex.org/C3019555358","wikidata":"https://www.wikidata.org/wiki/Q786423","display_name":"Fourier domain","level":3,"score":0.26820001006126404},{"id":"https://openalex.org/C108710211","wikidata":"https://www.wikidata.org/wiki/Q11538","display_name":"Mathematical proof","level":2,"score":0.2630999982357025}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1609/aaai.v40i34.40112","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i34.40112","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"}],"best_oa_location":{"id":"doi:10.1609/aaai.v40i34.40112","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i34.40112","pdf_url":null,"source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/7","score":0.7415018081665039,"display_name":"Affordable and clean energy"}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"The":[0],"emergence":[1],"of":[2,31,108],"accurate":[3],"open":[4],"large":[5],"language":[6],"models":[7],"(LLMs)":[8],"has":[9],"sparked":[10],"a":[11,42,50,133],"push":[12],"for":[13,37,154],"advanced":[14],"quantization":[15,36,98,153],"techniques":[16],"to":[17,74,88,164,168],"enable":[18,128],"efficient":[19],"deployment":[20],"on":[21,103,124,144],"end-user":[22],"devices.":[23],"In":[24,60,78],"this":[25],"paper,":[26],"we":[27,82,131],"revisit":[28],"the":[29,61,71,79,104,109,160],"challenge":[30],"extreme":[32],"LLM":[33],"compression---targeting":[34],"ultra-low-bit":[35],"both":[38,155],"activations":[39],"and":[40,57,68,157,176],"weights---from":[41],"Fourier":[43,86],"frequency":[44],"domain":[45],"perspective.":[46],"We":[47],"propose":[48],"SpecQuant,":[49],"two-stage":[51],"framework":[52],"that":[53,106,139],"tackles":[54],"activation":[55,64],"outliers":[56,65],"cross-channel":[58],"variance.":[59],"first":[62],"stage,":[63,81],"are":[66],"smoothed":[67],"transferred":[69],"into":[70],"weight":[72,110],"matrix":[73],"simplify":[75],"downstream":[76],"quantization.":[77],"second":[80],"apply":[83],"channel-wise":[84],"low-frequency":[85,115],"truncation":[87,135,141],"suppress":[89],"high-frequency":[90],"components":[91],"while":[92,171],"preserving":[93],"essential":[94],"signal":[95],"energy,":[96],"improving":[97],"robustness.":[99],"Our":[100],"method":[101],"builds":[102],"principle":[105],"most":[107],"energy":[111],"is":[112],"concentrated":[113],"in":[114],"components,":[116],"which":[117],"can":[118],"be":[119],"retained":[120],"with":[121],"minimal":[122],"impact":[123],"model":[125],"accuracy.":[126],"To":[127],"runtime":[129],"adaptability,":[130],"introduce":[132],"lightweight":[134],"module":[136],"during":[137],"inference":[138,175],"adjusts":[140],"thresholds":[142],"based":[143],"channel":[145],"characteristics.":[146],"On":[147],"LLaMA-3":[148],"8B,":[149],"SpecQuant":[150],"achieves":[151],"4-bit":[152],"weights":[156],"activations,":[158],"narrowing":[159],"zero-shot":[161],"accuracy":[162],"gap":[163],"only":[165],"1.5%":[166],"compared":[167],"full":[169],"precision,":[170],"delivering":[172],"2\u00d7":[173],"faster":[174],"3\u00d7":[177],"lower":[178],"memory":[179],"usage.":[180]},"counts_by_year":[],"updated_date":"2026-07-15T18:14:33.161393","created_date":"2026-03-18T00:00:00"}