SNAS: Fast Hardware-Aware Neural Architecture Search Methodology

Recently, automated neural architecture search (NAS) emerges as the default technique to find a state-of-the-art (SOTA) convolutional neural network (CNN) architecture with higher accuracy than manually designed architectures for image classification. In this article, we present a fast hardware-awar...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on computer-aided design of integrated circuits and systems Vol. 41; no. 11; pp. 4826 - 4836
Main Authors: Lee, Jaeseong, Rhim, Jungsub, Kang, Duseok, Ha, Soonhoi
Format: Journal Article
Language:English
Published: New York IEEE 01.11.2022
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Subjects:
ISSN:0278-0070, 1937-4151
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Abstract Recently, automated neural architecture search (NAS) emerges as the default technique to find a state-of-the-art (SOTA) convolutional neural network (CNN) architecture with higher accuracy than manually designed architectures for image classification. In this article, we present a fast hardware-aware NAS methodology, called S3NAS, reflecting the latest research results. It consists of three steps: 1) supernet design; 2) Single-Path NAS for fast architecture exploration; and 3) scaling and post-processing. In the first step, we design a supernet, superset of candidate networks with two features: one is to allow stages to have a different number of blocks, and the other is to enable blocks to have parallel layers of different kernel sizes (MixConv). Next, we perform a differential search by extending the Single-Path NAS technique to support the MixConv layer and to add a latency-aware loss term to reduce the hyperparameter search overhead. Finally, we use compound scaling to scale up the network maximally within the latency constraint. In addition, we add squeeze-and-excitation (SE) blocks and h-swish activation functions if beneficial in the post-processing step. Experiments with the proposed methodology on four different hardware platforms demonstrate the effectiveness of the proposed methodology. It is capable of finding networks with better latency-accuracy tradeoff than SOTA networks, and the network search can be done within 4 h using TPUv3.
AbstractList Recently, automated neural architecture search (NAS) emerges as the default technique to find a state-of-the-art (SOTA) convolutional neural network (CNN) architecture with higher accuracy than manually designed architectures for image classification. In this article, we present a fast hardware-aware NAS methodology, called S3NAS, reflecting the latest research results. It consists of three steps: 1) supernet design; 2) Single-Path NAS for fast architecture exploration; and 3) scaling and post-processing. In the first step, we design a supernet, superset of candidate networks with two features: one is to allow stages to have a different number of blocks, and the other is to enable blocks to have parallel layers of different kernel sizes (MixConv). Next, we perform a differential search by extending the Single-Path NAS technique to support the MixConv layer and to add a latency-aware loss term to reduce the hyperparameter search overhead. Finally, we use compound scaling to scale up the network maximally within the latency constraint. In addition, we add squeeze-and-excitation (SE) blocks and h-swish activation functions if beneficial in the post-processing step. Experiments with the proposed methodology on four different hardware platforms demonstrate the effectiveness of the proposed methodology. It is capable of finding networks with better latency-accuracy tradeoff than SOTA networks, and the network search can be done within 4 h using TPUv3.
Author Lee, Jaeseong
Kang, Duseok
Ha, Soonhoi
Rhim, Jungsub
Author_xml – sequence: 1
  givenname: Jaeseong
  orcidid: 0000-0002-3311-3891
  surname: Lee
  fullname: Lee, Jaeseong
  organization: Computer Science and Engineering Department, Seoul National University, Seoul, South Korea
– sequence: 2
  givenname: Jungsub
  surname: Rhim
  fullname: Rhim, Jungsub
  organization: Interdisciplinary Program in Artificial Intelligence, Seoul National University, Seoul, South Korea
– sequence: 3
  givenname: Duseok
  orcidid: 0000-0003-4985-0789
  surname: Kang
  fullname: Kang, Duseok
  organization: Computer Science and Engineering Department, Seoul National University, Seoul, South Korea
– sequence: 4
  givenname: Soonhoi
  orcidid: 0000-0001-7472-9142
  surname: Ha
  fullname: Ha, Soonhoi
  email: sha@snu.ac.kr
  organization: Computer Science and Engineering Department, Seoul National University, Seoul, South Korea
BookMark eNp9kMFOwzAMhiM0JLbBAyAulTh3xEnaJJyoBmNIYxw2zlWauqxTWUeaCu3tabWJAwcutmz58y99IzLY1Tsk5BroBIDqu_U0eZwwymDCgQsl-BkZguYyFBDBgAwpkyqkVNILMmqaLaUgIqaH5GG1TFb3wcw0Ppgbl38bh2HS12CJrTNVkDi7KT1a33a7FZpuDF7Rb-q8ruqPwyU5L0zV4NWpj8n77Gk9nYeLt-eXabIILaPKhxw0EzqnIlOxtZnKkRspmIqNAQMac82VEAXmmZaFKESMjBlLM4uZgNhKPia3x797V3-12Ph0W7du10WmTDIVSaARdFdwvLKubhqHRbp35adxhxRo2otKe1FpLyo9ieoY-YexpTe-rHfembL6l7w5kiUi_ibpWMRaR_wHJXx2aQ
CODEN ITCSDI
CitedBy_id crossref_primary_10_1145_3665868
crossref_primary_10_1145_3687310
crossref_primary_10_1109_ACCESS_2023_3286030
crossref_primary_10_1109_TCAD_2022_3223852
crossref_primary_10_1016_j_rser_2025_116231
crossref_primary_10_1109_TCSII_2023_3241487
crossref_primary_10_1109_TNNLS_2023_3346169
crossref_primary_10_1109_TNNLS_2025_3552693
crossref_primary_10_1109_TAI_2024_3377147
crossref_primary_10_1016_j_engappai_2025_110775
Cites_doi 10.1109/ACCESS.2020.3022327
10.1007/978-3-030-46147-8_29
10.1109/CVPR.2019.00293
10.1109/CVPR.2018.00474
10.1109/CVPR.2018.00907
10.1109/CVPR.2019.01099
10.1109/CVPR.2017.761
10.1145/3400302.3415731
10.1109/ETFA46521.2020.9212130
10.1109/JSTSP.2020.2971421
10.1109/CVPR.2017.668
10.1007/978-3-030-58539-6_26
10.1109/CVPR.2017.195
10.1109/TPAMI.2019.2913372
10.1609/aaai.v34i04.5959
10.1007/978-3-319-46493-0_39
10.1109/CVPR42600.2020.01044
10.1109/ICCD46524.2019.00102
10.1109/ICCV.2019.00140
ContentType Journal Article
Copyright Copyright The Institute of Electrical and Electronics Engineers, Inc. (IEEE) 2022
Copyright_xml – notice: Copyright The Institute of Electrical and Electronics Engineers, Inc. (IEEE) 2022
DBID 97E
RIA
RIE
AAYXX
CITATION
7SC
7SP
8FD
JQ2
L7M
L~C
L~D
DOI 10.1109/TCAD.2021.3134843
DatabaseName IEEE All-Society Periodicals Package (ASPP) 2005–Present
IEEE All-Society Periodicals Package (ASPP) 1998–Present
IEEE Xplore
CrossRef
Computer and Information Systems Abstracts
Electronics & Communications Abstracts
Technology Research Database
ProQuest Computer Science Collection
Advanced Technologies Database with Aerospace
Computer and Information Systems Abstracts – Academic
Computer and Information Systems Abstracts Professional
DatabaseTitle CrossRef
Technology Research Database
Computer and Information Systems Abstracts – Academic
Electronics & Communications Abstracts
ProQuest Computer Science Collection
Computer and Information Systems Abstracts
Advanced Technologies Database with Aerospace
Computer and Information Systems Abstracts Professional
DatabaseTitleList
Technology Research Database
Database_xml – sequence: 1
  dbid: RIE
  name: IEEE Xplore
  url: https://ieeexplore.ieee.org/
  sourceTypes: Publisher
DeliveryMethod fulltext_linktorsrc
Discipline Engineering
EISSN 1937-4151
EndPage 4836
ExternalDocumentID 10_1109_TCAD_2021_3134843
9646995
Genre orig-research
GrantInformation_xml – fundername: National Research Foundation of Korea (NRF)
  funderid: 10.13039/501100003725
– fundername: Korea Government (MSIT)
  grantid: NRF-2019R1A2B5B02069406
  funderid: 10.13039/501100003621
– fundername: TensorFlow Research Cloud (TFRC) Program
GroupedDBID --Z
-~X
0R~
29I
4.4
5GY
5VS
6IK
97E
AAJGR
AARMG
AASAJ
AAWTH
ABAZT
ABQJQ
ABVLG
ACGFS
ACIWK
ACNCT
AENEX
AETIX
AGQYO
AGSQL
AHBIQ
AI.
AIBXA
AKJIK
AKQYR
ALLEH
ALMA_UNASSIGNED_HOLDINGS
ASUFR
ATWAV
BEFXN
BFFAM
BGNUA
BKEBE
BPEOZ
CS3
DU5
EBS
EJD
HZ~
H~9
IBMZZ
ICLAB
IFIPE
IFJZH
IPLJI
JAVBF
LAI
M43
O9-
OCL
P2P
PZZ
RIA
RIE
RNS
TN5
VH1
VJK
AAYXX
CITATION
7SC
7SP
8FD
JQ2
L7M
L~C
L~D
ID FETCH-LOGICAL-c208t-319249d04b86ccb8de3a74286aa1a19ed93844fedb97f4f46e22ac0bceb416c73
IEDL.DBID RIE
ISSN 0278-0070
IngestDate Sun Nov 09 07:08:29 EST 2025
Sat Nov 29 03:31:49 EST 2025
Tue Nov 18 22:16:38 EST 2025
Wed Aug 27 02:14:20 EDT 2025
IsPeerReviewed true
IsScholarly true
Issue 11
Language English
License https://ieeexplore.ieee.org/Xplorehelp/downloads/license-information/IEEE.html
https://doi.org/10.15223/policy-029
https://doi.org/10.15223/policy-037
LinkModel DirectLink
MergedId FETCHMERGED-LOGICAL-c208t-319249d04b86ccb8de3a74286aa1a19ed93844fedb97f4f46e22ac0bceb416c73
Notes ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ORCID 0000-0003-4985-0789
0000-0001-7472-9142
0000-0002-3311-3891
PQID 2728571051
PQPubID 85470
PageCount 11
ParticipantIDs ieee_primary_9646995
crossref_primary_10_1109_TCAD_2021_3134843
crossref_citationtrail_10_1109_TCAD_2021_3134843
proquest_journals_2728571051
PublicationCentury 2000
PublicationDate 2022-11-01
PublicationDateYYYYMMDD 2022-11-01
PublicationDate_xml – month: 11
  year: 2022
  text: 2022-11-01
  day: 01
PublicationDecade 2020
PublicationPlace New York
PublicationPlace_xml – name: New York
PublicationTitle IEEE transactions on computer-aided design of integrated circuits and systems
PublicationTitleAbbrev TCAD
PublicationYear 2022
Publisher IEEE
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Publisher_xml – name: IEEE
– name: The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
References ref13
Liu (ref16)
ref15
ref14
ref31
ref30
ref11
ref10
ref2
Cai (ref22)
Li (ref1)
ref18
Xu (ref19)
ref24
ref23
ref26
ref20
Kim (ref3) 2015
Cai (ref17)
ref21
ref27
ref29
Gupta (ref6)
ref8
ref4
Mei (ref25)
ref5
Ramachandran (ref28)
Tan (ref7)
Zoph (ref12)
Tan (ref9)
References_xml – volume-title: Proc. ICLR
  ident: ref17
  article-title: ProxylessNAS: Direct neural architecture search on target task and hardware
– volume-title: Proc. ICLR
  ident: ref1
  article-title: Pruning filters for efficient ConvNets
– volume-title: Proc. MLSys Workshop
  ident: ref6
  article-title: Accelerator-aware neural network design using AutoML
– ident: ref21
  doi: 10.1109/ACCESS.2020.3022327
– ident: ref8
  doi: 10.1007/978-3-030-46147-8_29
– volume-title: Proc. ICLR
  ident: ref22
  article-title: Once-for-all: Train one network and specialize it for efficient deployment
– ident: ref14
  doi: 10.1109/CVPR.2019.00293
– ident: ref5
  doi: 10.1109/CVPR.2018.00474
– start-page: 74
  volume-title: Proc. BMVC
  ident: ref7
  article-title: MixConv: Mixed depthwise convolutional kernels
– ident: ref13
  doi: 10.1109/CVPR.2018.00907
– ident: ref18
  doi: 10.1109/CVPR.2019.01099
– ident: ref2
  doi: 10.1109/CVPR.2017.761
– ident: ref15
  doi: 10.1145/3400302.3415731
– ident: ref30
  doi: 10.1109/ETFA46521.2020.9212130
– ident: ref20
  doi: 10.1109/JSTSP.2020.2971421
– ident: ref26
  doi: 10.1109/CVPR.2017.668
– ident: ref29
  doi: 10.1007/978-3-030-58539-6_26
– ident: ref4
  doi: 10.1109/CVPR.2017.195
– ident: ref10
  doi: 10.1109/TPAMI.2019.2913372
– year: 2015
  ident: ref3
  article-title: Compression of deep convolutional neural networks for fast and low power mobile applications
  publication-title: arXiv:1511.06530
– volume-title: Proc. ICLR Workshop
  ident: ref28
  article-title: Searching for activation functions
– start-page: 6105
  volume-title: Proc. ICML
  ident: ref9
  article-title: EfficientNet: Rethinking model scaling for convolutional neural networks
– ident: ref23
  doi: 10.1609/aaai.v34i04.5959
– volume-title: Proc. Int. Conf. Learn. Represent.
  ident: ref19
  article-title: PC-DARTS: Partial channel connections for memory-efficient architecture search
– ident: ref31
  doi: 10.1007/978-3-319-46493-0_39
– ident: ref24
  doi: 10.1109/CVPR42600.2020.01044
– volume-title: Proc. ICLR
  ident: ref12
  article-title: Neural architecture search with reinforcement learning
– volume-title: Proc. ICLR
  ident: ref16
  article-title: DARTS: Differentiable architecture search
– ident: ref11
  doi: 10.1109/ICCD46524.2019.00102
– volume-title: Proc. ICLR
  ident: ref25
  article-title: AtomNAS: Fine-grained end-to-end neural architecture search
– ident: ref27
  doi: 10.1109/ICCV.2019.00140
SSID ssj0014529
Score 2.469639
Snippet Recently, automated neural architecture search (NAS) emerges as the default technique to find a state-of-the-art (SOTA) convolutional neural network (CNN)...
SourceID proquest
crossref
ieee
SourceType Aggregation Database
Enrichment Source
Index Database
Publisher
StartPage 4826
SubjectTerms Artificial neural networks
Compounds
Computer architecture
Constraint-aware AutoML
Convolution
Convolutional neural networks
convolutional neural networks (CNNs)
Hardware
Image classification
Kernel
Methodology
Network latency
neural architecture search (NAS)
neural network design
Neural networks
Searching
Space exploration
Title SNAS: Fast Hardware-Aware Neural Architecture Search Methodology
URI https://ieeexplore.ieee.org/document/9646995
https://www.proquest.com/docview/2728571051
Volume 41
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
journalDatabaseRights – providerCode: PRVIEE
  databaseName: IEEE Xplore
  customDbUrl:
  eissn: 1937-4151
  dateEnd: 99991231
  omitProxy: false
  ssIdentifier: ssj0014529
  issn: 0278-0070
  databaseCode: RIE
  dateStart: 19820101
  isFulltext: true
  titleUrlDefault: https://ieeexplore.ieee.org/
  providerName: IEEE
link http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV3dS8MwED-24YM--DXF6ZQ--CTWpWnaJD45xOGLQ9iEvZU0uYIgm2yd_vsmaTcVRfClpJCD9vJ1v_vl7gDOuY5zadIkVAZJyCzACBVaQ64oXJxPWrgkYr7YBB8OxWQiHxtwuY6FQUR_-QyvXNNz-Waml85V1pOpBXMyaUKT87SK1VozBo5A9P4UlzHWzuOawYyI7I3tT1kkSCMLUGMmWPztDPJFVX7sxP54Gez878N2Ybs2I4N-Ne570MDpPmx9SS7YhpvRsD-6DgZqUQaOn39Xcwz77hm4jBxe-pNECKp7x8GDryjtfe0H8DS4G9_eh3W9hFBTIkq7nTowZQjLRap1LgzGyiJfkSoVqUiikbFgrECTS16wgqVIqdIk15hbs0zz-BBa09kUjyCwq1IYTllOVMFitC-0oCh0SqQ2Oo46QFYazHSdTNzVtHjJPKggMnNKz5zSs1rpHbhYi7xWmTT-6tx2Wl53rBXcge5qmLJ6rS0yyqlIrKGURMe_S53AJnVBCz6CsAutcr7EU9jQb-XzYn7mp9EH0MPDQQ
linkProvider IEEE
linkToHtml http://cvtisr.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwlV3dS8MwED_mFNQHv6Y4ndoHn8S6NE3bxCeHOCZuQ9iEvZU0uYIgm-xD_32TrJuKIvhSUshBe_m63_1ydwDniQozoePIlxqJzwzA8CUaQy7PbZxPnNskYq7YRNLt8sFAPJbgchkLg4ju8hle2abj8vVIzayrrC5iA-ZEtAKrtnJWEa215Awsheg8KjZnrJnJBYcZEFHvm98yWJAGBqKGjLPw2ynkyqr82IvdAdPc_t-n7cBWYUh6jfnI70IJh3uw-SW9YAVuet1G79prysnUswz9uxyj37BPz-bkcNKfNII3v3nsdVxNaedt34en5l3_tuUXFRN8RQmfmg3VwilNWMZjpTKuMZQG-_JYykAGArUIOWM56kwkOctZjJRKRTKFmTHMVBIeQHk4GuIheGZdcp1QlhGZsxDNC80pchUTobQKgyqQhQZTVaQTt1UtXlIHK4hIrdJTq_S0UHoVLpYir_NcGn91rlgtLzsWCq5CbTFMabHaJilNKI-MqRQFR79LncF6q99pp-377sMxbFAbwuDiCWtQno5neAJr6m36PBmfuin1ASskxoo
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=SNAS%3A+Fast+Hardware-Aware+Neural+Architecture+Search+Methodology&rft.jtitle=IEEE+transactions+on+computer-aided+design+of+integrated+circuits+and+systems&rft.au=Lee%2C+Jaeseong&rft.au=Rhim%2C+Jungsub&rft.au=Kang%2C+Duseok&rft.au=Ha%2C+Soonhoi&rft.date=2022-11-01&rft.issn=0278-0070&rft.eissn=1937-4151&rft.volume=41&rft.issue=11&rft.spage=4826&rft.epage=4836&rft_id=info:doi/10.1109%2FTCAD.2021.3134843&rft.externalDBID=n%2Fa&rft.externalDocID=10_1109_TCAD_2021_3134843
thumbnail_l http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/lc.gif&issn=0278-0070&client=summon
thumbnail_m http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/mc.gif&issn=0278-0070&client=summon
thumbnail_s http://covers-cdn.summon.serialssolutions.com/index.aspx?isbn=/sc.gif&issn=0278-0070&client=summon