{"id":"https://openalex.org/W4391157567","doi":"https://doi.org/10.48550/arxiv.2401.11243","title":"LRP-QViT: Mixed-Precision Vision Transformer Quantization via Layer-wise Relevance Propagation","display_name":"LRP-QViT: Mixed-Precision Vision Transformer Quantization via Layer-wise Relevance Propagation","publication_year":2024,"publication_date":"2024-01-20","ids":{"openalex":"https://openalex.org/W4391157567","doi":"https://doi.org/10.48550/arxiv.2401.11243"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2401.11243","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2401.11243","pdf_url":"https://arxiv.org/pdf/2401.11243","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/2401.11243","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5029973259","display_name":"Navin Ranjan","orcid":"https://orcid.org/0000-0002-4608-0304","observed_orcids":["https://orcid.org/0000-0002-4608-0304"]},"institutions":[{"id":"https://openalex.org/I155173764","display_name":"Rochester Institute of Technology","ror":"https://ror.org/00v4yb702","country_code":"US","type":"education","lineage":["https://openalex.org/I155173764"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Ranjan, Navin","raw_affiliation_strings":["Rochester Institute of Technology Rochester , New York 14623 , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Rochester Institute of Technology Rochester , New York 14623 , USA","institution_ids":["https://openalex.org/I155173764"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5022041853","display_name":"Andreas E. Savakis","orcid":"https://orcid.org/0000-0002-9657-3027","observed_orcids":["https://orcid.org/0000-0002-9657-3027"]},"institutions":[{"id":"https://openalex.org/I155173764","display_name":"Rochester Institute of Technology","ror":"https://ror.org/00v4yb702","country_code":"US","type":"education","lineage":["https://openalex.org/I155173764"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Savakis, Andreas","raw_affiliation_strings":["Rochester Institute of Technology Rochester , New York 14623 , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Rochester Institute of Technology Rochester , New York 14623 , USA","institution_ids":["https://openalex.org/I155173764"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I155173764"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":17,"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":0.9940999746322632,"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":0.9940999746322632,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.004000000189989805,"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":0.0003000000142492354,"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/mixed-precision-quantization","display_name":"mixed-precision quantization","score":0.878000020980835},{"id":"https://openalex.org/keywords/vision-transformer","display_name":"vision transformer","score":0.906000018119812},{"id":"https://openalex.org/keywords/layer-wise-relevance-propagation","display_name":"layer-wise relevance propagation","score":0.9710000157356262},{"id":"https://openalex.org/keywords/bit-allocation","display_name":"bit allocation","score":0.6169999837875366},{"id":"https://openalex.org/keywords/post-training-quantization","display_name":"post-training quantization","score":0.800000011920929},{"id":"https://openalex.org/keywords/explainability","display_name":"explainability","score":0.5379999876022339},{"id":"https://openalex.org/keywords/model-compression","display_name":"model compression","score":0.675000011920929},{"id":"https://openalex.org/keywords/deit","display_name":"DeiT","score":0.6129999756813049},{"id":"https://openalex.org/keywords/swin-transformer","display_name":"Swin Transformer","score":0.9850000143051147}],"concepts":[{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.7417198419570923},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7215351462364197},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5391059517860413},{"id":"https://openalex.org/C79337645","wikidata":"https://www.wikidata.org/wiki/Q779824","display_name":"Outlier","level":2,"score":0.5022974014282227},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.5001530647277832},{"id":"https://openalex.org/C158154518","wikidata":"https://www.wikidata.org/wiki/Q7310970","display_name":"Relevance (law)","level":2,"score":0.433779776096344},{"id":"https://openalex.org/C113775141","wikidata":"https://www.wikidata.org/wiki/Q428691","display_name":"Computer engineering","level":1,"score":0.38699033856391907},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.35259944200515747},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.32184016704559326},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.08734524250030518},{"id":"https://openalex.org/C199539241","wikidata":"https://www.wikidata.org/wiki/Q7748","display_name":"Law","level":1,"score":0.0},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.0},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.0},{"id":"https://openalex.org/C17744445","wikidata":"https://www.wikidata.org/wiki/Q36442","display_name":"Political science","level":0,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2401.11243","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2401.11243","pdf_url":"https://arxiv.org/pdf/2401.11243","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.2401.11243","doi":"https://doi.org/10.48550/arxiv.2401.11243","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2401.11243","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:2401.11243","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2401.11243","pdf_url":"https://arxiv.org/pdf/2401.11243","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":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4391157567.pdf","grobid_xml":"https://content.openalex.org/works/W4391157567.grobid-xml"},"referenced_works_count":4,"referenced_works":["https://openalex.org/W2626639386","https://openalex.org/W2899663614","https://openalex.org/W3022265721","https://openalex.org/W4385484624"],"related_works":["https://openalex.org/W4412445072","https://openalex.org/W6910584288","https://openalex.org/W4416238258","https://openalex.org/W7169656317","https://openalex.org/W4391335338","https://openalex.org/W7171938145","https://openalex.org/W4385484624","https://openalex.org/W4414093498","https://openalex.org/W7155364465","https://openalex.org/W7212136963"],"abstract_inverted_index":{"Vision":[0],"transformers":[1],"(ViTs)":[2],"have":[3,162],"demonstrated":[4],"remarkable":[5],"performance":[6],"across":[7,189],"various":[8,197],"visual":[9],"tasks.":[10],"However,":[11],"ViT":[12],"models":[13,195,214],"suffer":[14],"from":[15,172],"substantial":[16],"computational":[17],"and":[18,133,182,192,207,220],"memory":[19],"requirements,":[20],"making":[21],"it":[22,135,143],"challenging":[23],"to":[24,94,104,175],"deploy":[25],"them":[26],"on":[27,44,61,71,98,196],"resource-constrained":[28],"platforms.":[29],"Quantization":[30],"is":[31],"a":[32,164],"popular":[33],"approach":[34],"for":[35,48,66,89,154],"reducing":[36],"model":[37],"size,":[38],"but":[39],"most":[40],"studies":[41],"mainly":[42],"focus":[43],"equal":[45],"bit-width":[46],"quantization":[47,64,167,210],"the":[49,106,114,119,130,140,145,156,216],"entire":[50],"network,":[51],"resulting":[52],"in":[53,112,215],"sub-optimal":[54],"solutions.":[55],"While":[56],"there":[57],"are":[58],"few":[59],"works":[60],"mixed":[62,77],"precision":[63,78],"(MPQ)":[65],"ViTs,":[67],"they":[68],"typically":[69],"rely":[70],"search":[72],"space-based":[73],"methods":[74,211],"or":[75],"employ":[76,118,187],"arbitrarily.":[79],"In":[80],"this":[81],"paper,":[82],"we":[83,117,161,186],"introduce":[84],"LRP-QViT,":[85],"an":[86],"explainability-based":[87],"method":[88],"assigning":[90],"mixed-precision":[91],"bit":[92],"allocations":[93],"different":[95],"layers":[96],"based":[97],"their":[99],"importance":[100],"during":[101],"classification.":[102],"Specifically,":[103],"measure":[105],"contribution":[107,158],"score":[108],"of":[109,218],"each":[110],"layer":[111,132,157],"predicting":[113],"target":[115],"class,":[116],"Layer-wise":[120],"Relevance":[121],"Propagation":[122],"(LRP)":[123],"method.":[124],"LRP":[125],"assigns":[126],"local":[127],"relevance":[128,141,149],"at":[129,169],"output":[131],"propagates":[134],"through":[136],"all":[137],"layers,":[138],"distributing":[139],"until":[142],"reaches":[144],"input":[146],"layers.":[147],"These":[148],"scores":[150],"serve":[151],"as":[152],"indicators":[153],"computing":[155],"score.":[159],"Additionally,":[160],"introduced":[163],"clipped":[165],"channel-wise":[166],"aimed":[168],"eliminating":[170],"outliers":[171],"post-LayerNorm":[173],"activations":[174],"alleviate":[176],"severe":[177],"inter-channel":[178],"variations.":[179],"To":[180],"validate":[181],"assess":[183],"our":[184,205],"approach,":[185],"LRP-QViT":[188],"ViT,":[190],"DeiT,":[191],"Swin":[193],"transformer":[194],"datasets.":[198],"Our":[199],"experimental":[200],"findings":[201],"demonstrate":[202],"that":[203],"both":[204],"fixed-bit":[206],"mixed-bit":[208],"post-training":[209],"surpass":[212],"existing":[213],"context":[217],"4-bit":[219],"6-bit":[221],"quantization.":[222]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":9},{"year":2024,"cited_by_count":4},{"year":2023,"cited_by_count":1}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2024-01-24T00:00:00"}