{"id":"https://openalex.org/W4416341681","doi":"https://doi.org/10.1109/socc66126.2025.11235478","title":"Exploring Training of Approximate CNNs using Straight-Through and Linear Regression-Based Gradient Estimation","display_name":"Exploring Training of Approximate CNNs using Straight-Through and Linear Regression-Based Gradient Estimation","publication_year":2025,"publication_date":"2025-09-29","ids":{"openalex":"https://openalex.org/W4416341681","doi":"https://doi.org/10.1109/socc66126.2025.11235478"},"language":null,"primary_location":{"id":"doi:10.1109/socc66126.2025.11235478","is_oa":false,"landing_page_url":"https://doi.org/10.1109/socc66126.2025.11235478","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE 38th International System-on-Chip Conference (SOCC)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":false,"oa_status":"closed","oa_url":null,"any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5100664932","display_name":"Ke Ma","orcid":"https://orcid.org/0000-0001-8853-7782"},"institutions":[{"id":"https://openalex.org/I150744194","display_name":"Waseda University","ror":"https://ror.org/00ntfnx83","country_code":"JP","type":"education","lineage":["https://openalex.org/I150744194"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Ke Ma","raw_affiliation_strings":["Waseda University,Graduate School of Information, Production and Systems,Kitakyushu,Japan"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Waseda University,Graduate School of Information, Production and Systems,Kitakyushu,Japan","institution_ids":["https://openalex.org/I150744194"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5078160965","display_name":"Mingtao Zhang","orcid":"https://orcid.org/0000-0002-6814-4962"},"institutions":[{"id":"https://openalex.org/I22299242","display_name":"Kyoto University","ror":"https://ror.org/02kpeqv85","country_code":"JP","type":"education","lineage":["https://openalex.org/I22299242"]},{"id":"https://openalex.org/I39012071","display_name":"Kyoto College of Graduate Studies for Informatics","ror":"https://ror.org/05mzj8a56","country_code":"JP","type":"education","lineage":["https://openalex.org/I39012071"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Mingtao Zhang","raw_affiliation_strings":["Kyoto University,Graduate School of Informatics,Kyoto,Japan"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Kyoto University,Graduate School of Informatics,Kyoto,Japan","institution_ids":["https://openalex.org/I39012071","https://openalex.org/I22299242"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5042142715","display_name":"Shinichi Nishizawa","orcid":"https://orcid.org/0000-0002-1172-7286"},"institutions":[{"id":"https://openalex.org/I113306721","display_name":"Hiroshima University","ror":"https://ror.org/03t78wx29","country_code":"JP","type":"education","lineage":["https://openalex.org/I113306721"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Shinichi Nishizawa","raw_affiliation_strings":["Hiroshima University,Graduate School of Advanced Science and Engineering,Hiroshima,Japan"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Hiroshima University,Graduate School of Advanced Science and Engineering,Hiroshima,Japan","institution_ids":["https://openalex.org/I113306721"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5103496033","display_name":"Shinji Kimura","orcid":null},"institutions":[{"id":"https://openalex.org/I150744194","display_name":"Waseda University","ror":"https://ror.org/00ntfnx83","country_code":"JP","type":"education","lineage":["https://openalex.org/I150744194"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Shinji Kimura","raw_affiliation_strings":["Waseda University,Graduate School of Information, Production and Systems,Kitakyushu,Japan"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Waseda University,Graduate School of Information, Production and Systems,Kitakyushu,Japan","institution_ids":["https://openalex.org/I150744194"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"1","last_page":"6"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10363","display_name":"Low-power high-performance VLSI design","score":0.4521999955177307,"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"}},"topics":[{"id":"https://openalex.org/T10363","display_name":"Low-power high-performance VLSI design","score":0.4521999955177307,"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/T10036","display_name":"Advanced Neural Network Applications","score":0.2849999964237213,"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/T11697","display_name":"Numerical Methods and Algorithms","score":0.07209999859333038,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/estimator","display_name":"Estimator","score":0.7305999994277954},{"id":"https://openalex.org/keywords/quantization","display_name":"Quantization (signal processing)","score":0.6913999915122986},{"id":"https://openalex.org/keywords/inference","display_name":"Inference","score":0.6176000237464905},{"id":"https://openalex.org/keywords/convolutional-neural-network","display_name":"Convolutional neural network","score":0.6011000275611877},{"id":"https://openalex.org/keywords/gradient-descent","display_name":"Gradient descent","score":0.5670999884605408},{"id":"https://openalex.org/keywords/multiplier","display_name":"Multiplier (economics)","score":0.5131999850273132},{"id":"https://openalex.org/keywords/artificial-neural-network","display_name":"Artificial neural network","score":0.4702000021934509},{"id":"https://openalex.org/keywords/convergence","display_name":"Convergence (economics)","score":0.44679999351501465}],"concepts":[{"id":"https://openalex.org/C185429906","wikidata":"https://www.wikidata.org/wiki/Q1130160","display_name":"Estimator","level":2,"score":0.7305999994277954},{"id":"https://openalex.org/C28855332","wikidata":"https://www.wikidata.org/wiki/Q198099","display_name":"Quantization (signal processing)","level":2,"score":0.6913999915122986},{"id":"https://openalex.org/C2776214188","wikidata":"https://www.wikidata.org/wiki/Q408386","display_name":"Inference","level":2,"score":0.6176000237464905},{"id":"https://openalex.org/C81363708","wikidata":"https://www.wikidata.org/wiki/Q17084460","display_name":"Convolutional neural network","level":2,"score":0.6011000275611877},{"id":"https://openalex.org/C153258448","wikidata":"https://www.wikidata.org/wiki/Q1199743","display_name":"Gradient descent","level":3,"score":0.5670999884605408},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5419999957084656},{"id":"https://openalex.org/C124584101","wikidata":"https://www.wikidata.org/wiki/Q1053266","display_name":"Multiplier (economics)","level":2,"score":0.5131999850273132},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.48910000920295715},{"id":"https://openalex.org/C50644808","wikidata":"https://www.wikidata.org/wiki/Q192776","display_name":"Artificial neural network","level":2,"score":0.4702000021934509},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.4609000086784363},{"id":"https://openalex.org/C2777303404","wikidata":"https://www.wikidata.org/wiki/Q759757","display_name":"Convergence (economics)","level":2,"score":0.44679999351501465},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.43950000405311584},{"id":"https://openalex.org/C2777211547","wikidata":"https://www.wikidata.org/wiki/Q17141490","display_name":"Training (meteorology)","level":2,"score":0.35359999537467957},{"id":"https://openalex.org/C2984842247","wikidata":"https://www.wikidata.org/wiki/Q197536","display_name":"Deep neural networks","level":3,"score":0.3499999940395355},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.34929999709129333},{"id":"https://openalex.org/C151376022","wikidata":"https://www.wikidata.org/wiki/Q168698","display_name":"Exponential function","level":2,"score":0.3463999927043915},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.3450999855995178},{"id":"https://openalex.org/C2944601119","wikidata":"https://www.wikidata.org/wiki/Q43744058","display_name":"Residual neural network","level":3,"score":0.33980000019073486},{"id":"https://openalex.org/C163175372","wikidata":"https://www.wikidata.org/wiki/Q3339222","display_name":"Linear model","level":2,"score":0.3059000074863434},{"id":"https://openalex.org/C115680565","wikidata":"https://www.wikidata.org/wiki/Q5977448","display_name":"Gradient method","level":2,"score":0.2904999852180481},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.27880001068115234},{"id":"https://openalex.org/C139945424","wikidata":"https://www.wikidata.org/wiki/Q1940696","display_name":"Mean squared error","level":2,"score":0.2718000113964081},{"id":"https://openalex.org/C167928553","wikidata":"https://www.wikidata.org/wiki/Q1376021","display_name":"Estimation theory","level":2,"score":0.26429998874664307}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1109/socc66126.2025.11235478","is_oa":false,"landing_page_url":"https://doi.org/10.1109/socc66126.2025.11235478","pdf_url":null,"source":null,"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"2025 IEEE 38th International System-on-Chip Conference (SOCC)","raw_type":"proceedings-article"}],"best_oa_location":null,"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":false},"content_urls":null,"referenced_works_count":21,"referenced_works":["https://openalex.org/W2108598243","https://openalex.org/W2120000030","https://openalex.org/W2194775991","https://openalex.org/W2395611524","https://openalex.org/W2589986623","https://openalex.org/W2612139336","https://openalex.org/W2790285612","https://openalex.org/W2799723702","https://openalex.org/W2963122961","https://openalex.org/W3007372856","https://openalex.org/W3036274717","https://openalex.org/W3037251238","https://openalex.org/W4206350167","https://openalex.org/W4249697915","https://openalex.org/W4304693398","https://openalex.org/W4312712198","https://openalex.org/W4319431599","https://openalex.org/W4323021809","https://openalex.org/W4323345742","https://openalex.org/W4386952673","https://openalex.org/W4400235168"],"related_works":[],"abstract_inverted_index":{"With":[0],"the":[1,32,48,71,81,89,104,122,152,174,186],"rapid":[2],"growth":[3],"of":[4,34,50,73,91],"deep":[5],"convolutional":[6],"neural":[7],"network":[8],"(CNN)":[9],"architectures,":[10],"reducing":[11],"inference":[12,45],"cost":[13],"on":[14,44,134,148,160,169],"power-constrained":[15],"devices":[16],"has":[17,53],"become":[18],"critical.":[19],"Approximate":[20],"multipliers":[21],"(AMs)":[22],"offer":[23],"significant":[24],"savings":[25],"in":[26],"area,":[27],"power,":[28],"and":[29,47,76,103,151],"delay":[30],"at":[31],"expense":[33],"occasional":[35],"arithmetic":[36],"errors.":[37],"However,":[38],"existing":[39],"works":[40,158],"have":[41],"largely":[42],"focused":[43],"only,":[46],"training":[49,63,90],"AM-based":[51,64,92],"CNNs":[52,93],"not":[54],"been":[55],"explored":[56],"thoroughly.":[57],"Two":[58],"major":[59],"questions":[60],"exist":[61],"for":[62,138],"CNNs:":[65],"1)":[66],"how":[67,78],"to":[68,79,120,178],"accurately":[69],"estimate":[70],"gradients":[72],"approximate":[74],"multipliers,":[75],"2)":[77],"decide":[80],"quantization":[82,123,154],"scale.":[83],"In":[84],"this":[85],"paper,":[86],"we":[87,171],"explore":[88],"with":[94],"two":[95],"different":[96,128],"gradient":[97,108,162],"estimation":[98,163],"methods:":[99],"straight-through":[100],"estimator":[101,109],"(STE)":[102],"proposed":[105,141],"linear":[106],"regression-based":[107],"(LRE).":[110],"A":[111],"method":[112,157],"using":[113],"exponential":[114],"moving":[115],"average":[116],"(EMA)":[117],"is":[118,181],"tested":[119],"update":[121,156],"scale":[124,155],"during":[125],"training.":[126],"Seventeen":[127],"8-bit":[129],"signed":[130],"AMs":[131],"are":[132],"trained":[133],"three":[135],"ResNet":[136],"models":[137],"CIFAR-10.":[139],"The":[140],"LRE":[142],"shows":[143],"better":[144],"convergence":[145],"than":[146,185],"STE":[147],"specific":[149],"AMs,":[150],"EMA-based":[153],"well":[159],"both":[161],"methods.":[164],"For":[165],"multiplier":[166],"mul8s":[167],"1KXF":[168],"ResNet-18,":[170],"can":[172],"recover":[173],"accuracy":[175],"from":[176],"10.8%":[177],"94.31%":[179],"which":[180],"only":[182],"0.89%":[183],"lower":[184],"exact":[187],"floating-point":[188],"model.":[189]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-11-17T00:00:00"}