{"id":"https://openalex.org/W2890616119","doi":"https://doi.org/10.48550/arxiv.1809.04191","title":"Discovering Low-Precision Networks Close to Full-Precision Networks for Efficient Embedded Inference","display_name":"Discovering Low-Precision Networks Close to Full-Precision Networks for Efficient Embedded Inference","publication_year":2018,"publication_date":"2018-09-11","ids":{"openalex":"https://openalex.org/W2890616119","doi":"https://doi.org/10.48550/arxiv.1809.04191","mag":"2890616119"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:1809.04191","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/1809.04191","pdf_url":"https://arxiv.org/pdf/1809.04191","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/1809.04191","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5059206462","display_name":"Jeffrey L. McKinstry","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210085935","display_name":"IBM Research - Almaden","ror":"https://ror.org/005w8dd04","country_code":"US","type":"facility","lineage":["https://openalex.org/I1341412227","https://openalex.org/I4210085935","https://openalex.org/I4210114115"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"McKinstry, Jeffrey L.","raw_affiliation_strings":["IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA","institution_ids":["https://openalex.org/I4210085935"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5055044754","display_name":"Steven K. Esser","orcid":"https://orcid.org/0009-0008-4089-0577","observed_orcids":["https://orcid.org/0009-0008-4089-0577"]},"institutions":[{"id":"https://openalex.org/I4210085935","display_name":"IBM Research - Almaden","ror":"https://ror.org/005w8dd04","country_code":"US","type":"facility","lineage":["https://openalex.org/I1341412227","https://openalex.org/I4210085935","https://openalex.org/I4210114115"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Esser, Steven K.","raw_affiliation_strings":["IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA","institution_ids":["https://openalex.org/I4210085935"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5113665253","display_name":"Rathinakumar Appuswamy","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210085935","display_name":"IBM Research - Almaden","ror":"https://ror.org/005w8dd04","country_code":"US","type":"facility","lineage":["https://openalex.org/I1341412227","https://openalex.org/I4210085935","https://openalex.org/I4210114115"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Appuswamy, Rathinakumar","raw_affiliation_strings":["IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA","institution_ids":["https://openalex.org/I4210085935"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5045970662","display_name":"Deepika Bablani","orcid":"https://orcid.org/0000-0002-8987-8201","observed_orcids":["https://orcid.org/0000-0002-8987-8201"]},"institutions":[{"id":"https://openalex.org/I4210085935","display_name":"IBM Research - Almaden","ror":"https://ror.org/005w8dd04","country_code":"US","type":"facility","lineage":["https://openalex.org/I1341412227","https://openalex.org/I4210085935","https://openalex.org/I4210114115"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Bablani, Deepika","raw_affiliation_strings":["IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA","institution_ids":["https://openalex.org/I4210085935"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5087078947","display_name":"John V. Arthur","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210085935","display_name":"IBM Research - Almaden","ror":"https://ror.org/005w8dd04","country_code":"US","type":"facility","lineage":["https://openalex.org/I1341412227","https://openalex.org/I4210085935","https://openalex.org/I4210114115"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Arthur, John V.","raw_affiliation_strings":["IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA","institution_ids":["https://openalex.org/I4210085935"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5087595691","display_name":"Izzet B. Yildiz","orcid":"https://orcid.org/0000-0002-7998-4449","observed_orcids":["https://orcid.org/0000-0002-7998-4449"]},"institutions":[{"id":"https://openalex.org/I4210085935","display_name":"IBM Research - Almaden","ror":"https://ror.org/005w8dd04","country_code":"US","type":"facility","lineage":["https://openalex.org/I1341412227","https://openalex.org/I4210085935","https://openalex.org/I4210114115"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yildiz, Izzet B.","raw_affiliation_strings":["IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA","institution_ids":["https://openalex.org/I4210085935"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5063718511","display_name":"Dharmendra S. Modha","orcid":"https://orcid.org/0009-0005-6714-8127","observed_orcids":["https://orcid.org/0009-0005-6714-8127"]},"institutions":[{"id":"https://openalex.org/I4210085935","display_name":"IBM Research - Almaden","ror":"https://ror.org/005w8dd04","country_code":"US","type":"facility","lineage":["https://openalex.org/I1341412227","https://openalex.org/I4210085935","https://openalex.org/I4210114115"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Modha, Dharmendra S.","raw_affiliation_strings":["IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA","IBM"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"IBM Almaden Research Center 650 Harry Road , San Jose , CA 95120 , USA","institution_ids":["https://openalex.org/I4210085935"]},{"raw_affiliation_string":"IBM","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I4210085935"],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":96,"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":0.9945999979972839,"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":0.9945999979972839,"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/T10363","display_name":"Low-power high-performance VLSI design","score":0.0010999999940395355,"subfield":{"id":"https://openalex.org/subfields/2208","display_name":"Electrical and Electronic Engineering"},"field":{"id":"https://openalex.org/fields/22","display_name":"Engineering"},"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":0.0006000000284984708,"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/low-precision-neural-networks","display_name":"low-precision neural networks","score":0.8690000176429749},{"id":"https://openalex.org/keywords/4-bit-quantization","display_name":"4-bit quantization","score":0.7940000295639038},{"id":"https://openalex.org/keywords/8-bit-quantization","display_name":"8-bit quantization","score":0.9570000171661377},{"id":"https://openalex.org/keywords/imagenet-classification","display_name":"ImageNet classification","score":0.7459999918937683},{"id":"https://openalex.org/keywords/fine-tuning-pre-trained-models","display_name":"pre-trained model fine-tuning","score":0.8799999952316284},{"id":"https://openalex.org/keywords/gradient-noise","display_name":"gradient noise","score":0.7429999709129333},{"id":"https://openalex.org/keywords/quantization-noise","display_name":"quantization noise","score":0.5289999842643738},{"id":"https://openalex.org/keywords/resnet","display_name":"ResNet","score":0.6389999985694885},{"id":"https://openalex.org/keywords/inception-v3","display_name":"InceptionV3","score":0.890999972820282},{"id":"https://openalex.org/keywords/densenet-161","display_name":"DenseNet-161","score":0.8740000128746033}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7326486110687256},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.6126410365104675},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6047005653381348},{"id":"https://openalex.org/C185798385","wikidata":"https://www.wikidata.org/wiki/Q1161707","display_name":"Benchmark (surveying)","level":2,"score":0.5670033097267151},{"id":"https://openalex.org/C12725497","wikidata":"https://www.wikidata.org/wiki/Q810247","display_name":"Baseline (sea)","level":2,"score":0.5501280426979065},{"id":"https://openalex.org/C202799725","wikidata":"https://www.wikidata.org/wiki/Q272035","display_name":"Accuracy and precision","level":2,"score":0.5459288358688354},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4480849802494049},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.4161968231201172},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.38629287481307983},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.17580050230026245},{"id":"https://openalex.org/C105795698","wikidata":"https://www.wikidata.org/wiki/Q12483","display_name":"Statistics","level":1,"score":0.1499139368534088},{"id":"https://openalex.org/C127313418","wikidata":"https://www.wikidata.org/wiki/Q1069","display_name":"Geology","level":0,"score":0.0},{"id":"https://openalex.org/C13280743","wikidata":"https://www.wikidata.org/wiki/Q131089","display_name":"Geodesy","level":1,"score":0.0},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.0},{"id":"https://openalex.org/C115961682","wikidata":"https://www.wikidata.org/wiki/Q860623","display_name":"Image (mathematics)","level":2,"score":0.0},{"id":"https://openalex.org/C111368507","wikidata":"https://www.wikidata.org/wiki/Q43518","display_name":"Oceanography","level":1,"score":0.0}],"mesh":[],"locations_count":3,"locations":[{"id":"pmh:oai:arXiv.org:1809.04191","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/1809.04191","pdf_url":"https://arxiv.org/pdf/1809.04191","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":"mag:2890616119","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/pdf/1809.04191.pdf","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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"arXiv (Cornell University)","raw_type":null},{"id":"doi:10.48550/arxiv.1809.04191","doi":"https://doi.org/10.48550/arxiv.1809.04191","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.1809.04191","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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:1809.04191","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/1809.04191","pdf_url":"https://arxiv.org/pdf/1809.04191","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/W2890616119.pdf","grobid_xml":"https://content.openalex.org/works/W2890616119.grobid-xml"},"referenced_works_count":29,"referenced_works":["https://openalex.org/W1686810756","https://openalex.org/W1850240193","https://openalex.org/W2183341477","https://openalex.org/W2194775991","https://openalex.org/W2242818861","https://openalex.org/W2260663238","https://openalex.org/W2291160084","https://openalex.org/W2300242332","https://openalex.org/W2314470091","https://openalex.org/W2586654419","https://openalex.org/W2593245696","https://openalex.org/W2751477244","https://openalex.org/W2766164908","https://openalex.org/W2769312802","https://openalex.org/W2777406049","https://openalex.org/W2786771851","https://openalex.org/W2787513570","https://openalex.org/W2787752464","https://openalex.org/W2789142637","https://openalex.org/W2791110811","https://openalex.org/W2792287754","https://openalex.org/W2803549871","https://openalex.org/W2885688430","https://openalex.org/W2954582488","https://openalex.org/W2962958489","https://openalex.org/W2963114950","https://openalex.org/W2963446712","https://openalex.org/W2963521187","https://openalex.org/W6929263688"],"related_works":["https://openalex.org/W3173567919","https://openalex.org/W2947469857","https://openalex.org/W2593245696","https://openalex.org/W2994840239","https://openalex.org/W2890248397","https://openalex.org/W3027524057","https://openalex.org/W2962728029","https://openalex.org/W3165976180","https://openalex.org/W4317526202","https://openalex.org/W2584314014"],"abstract_inverted_index":{"To":[0,21],"realize":[1],"the":[2,38,55,64,67,78,97,100,105,111,114,124,127,173,177,180,184,188,192,245],"promise":[3,28],"of":[4,16,66,74,80,99,113,126,159,175,179,191,244],"ubiquitous":[5],"embedded":[6],"deep":[7],"network":[8],"inference,":[9],"it":[10],"is":[11,170],"essential":[12],"to":[13,108,123,142,153,164,172,240,250,258],"seek":[14,151],"limits":[15],"energy":[17,31],"and":[18,32,51,91,150,209,211,222],"area":[19,33],"efficiency.":[20],"this":[22,155],"end,":[23],"low-precision":[24,115,252],"networks":[25,53,70,103,116,208],"offer":[26],"tremendous":[27],"because":[29],"both":[30],"scale":[34],"down":[35],"quadratically":[36],"with":[37,147,203,234],"reduction":[39],"in":[40],"precision.":[41],"Here":[42],"we":[43,196],"demonstrate":[44,85],"ResNet-18,":[45,86],"-34,":[46,87],"-50,":[47,88],"-152,":[48,89],"Inception-v3,":[49],"Densenet-161,":[50,90],"VGG-16bn":[52,92],"on":[54],"ImageNet":[56],"classification":[57],"benchmark":[58],"that,":[59],"at":[60],"8-bit":[61],"precision":[62,206,260],"exceed":[63],"accuracy":[65,98],"full-precision":[68,101],"baseline":[69,102,129,207,261],"after":[71],"one":[72],"epoch":[73],"finetuning,":[75],"thereby":[76],"leveraging":[77],"availability":[79],"pretrained":[81,204],"models.":[82],"We":[83,136],"also":[84],"4-bit":[93],"models":[94],"that":[95,138,229,268],"match":[96],"--":[104],"highest":[106],"scores":[107],"date.":[109],"Surprisingly,":[110],"weights":[112,125],"are":[117,248],"very":[118],"close":[119,257],"(in":[120],"cosine":[121],"similarity)":[122],"corresponding":[128],"networks,":[130,253],"making":[131],"training":[132,145,168,220],"from":[133,183],"scratch":[134],"unnecessary.":[135],"find":[137],"gradient":[139,193,214],"noise":[140,215],"due":[141],"quantization":[143,218],"during":[144],"increases":[146],"reduced":[148],"precision,":[149,247],"ways":[152],"overcome":[154],"noise.":[156],"The":[157,263],"number":[158],"iterations":[160],"required":[161],"by":[162,201,217,219],"SGD":[163],"achieve":[165],"a":[166],"given":[167],"error":[169],"related":[171],"square":[174],"(a)":[176,197],"distance":[178,200],"initial":[181],"solution":[182,199],"final":[185],"plus":[186],"(b)":[187,212],"maximum":[189],"variance":[190],"estimates.":[194],"Therefore,":[195],"reduce":[198],"starting":[202],"fp32":[205,259],"fine-tuning,":[210],"combat":[213],"introduced":[216],"longer":[221],"reducing":[223],"learning":[224],"rates.":[225],"Sensitivity":[226],"analysis":[227],"indicates":[228],"these":[230],"simple":[231],"techniques,":[232],"coupled":[233],"proper":[235],"activation":[236],"function":[237],"range":[238],"calibration":[239],"take":[241],"full":[242],"advantage":[243],"limited":[246],"sufficient":[249],"discover":[251],"if":[254],"they":[255],"exist,":[256],"networks.":[262],"results":[264],"herein":[265],"provide":[266],"evidence":[267],"4-bits":[269],"suffice":[270],"for":[271],"classification.":[272]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":8},{"year":2024,"cited_by_count":8},{"year":2023,"cited_by_count":12},{"year":2022,"cited_by_count":13},{"year":2021,"cited_by_count":18},{"year":2020,"cited_by_count":17},{"year":2019,"cited_by_count":14},{"year":2018,"cited_by_count":3}],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2018-09-27T00:00:00"}