FrostNet: Towards Quantization-Aware Network Architecture Search

INT8 quantization has become one of the standard techniques for deploying convolutional neural networks (CNNs) on edge devices to reduce the memory and computational resource usages. By analyzing quantized performances of existing mobile-target network architectures, we can raise an issue regarding...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Kim, Taehoon, Yoo, YoungJoon, Yang, Jihoon
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Kim, Taehoon
Yoo, YoungJoon
Yang, Jihoon
description INT8 quantization has become one of the standard techniques for deploying convolutional neural networks (CNNs) on edge devices to reduce the memory and computational resource usages. By analyzing quantized performances of existing mobile-target network architectures, we can raise an issue regarding the importance of network architecture for optimal INT8 quantization. In this paper, we present a new network architecture search (NAS) procedure to find a network that guarantees both full-precision (FLOAT32) and quantized (INT8) performances. We first propose critical but straightforward optimization method which enables quantization-aware training (QAT) : floating-point statistic assisting (StatAssist) and stochastic gradient boosting (GradBoost). By integrating the gradient-based NAS with StatAssist and GradBoost, we discovered a quantization-efficient network building block, Frost bottleneck. Furthermore, we used Frost bottleneck as the building block for hardware-aware NAS to obtain quantization-efficient networks, FrostNets, which show improved quantization performances compared to other mobile-target networks while maintaining competitive FLOAT32 performance. Our FrostNets achieve higher recognition accuracy than existing CNNs with comparable latency when quantized, due to higher latency reduction rate (average 65%).
doi_str_mv 10.48550/arxiv.2006.09679
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2006_09679</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2006_09679</sourcerecordid><originalsourceid>FETCH-LOGICAL-a679-c83a56c114a01233e3fcc040ddade6d504ad3b8334d115f89e5bf5a83426ed643</originalsourceid><addsrcrecordid>eNotj8FOwzAQRH3hgAofwKn5gYR11jYOJ6KKAlIFQs092tobYQENclwKfD1u4TSaGWlGT4gLCZWyWsMlxa_wWdUApoLGXDWn4mYZxyk9crouunFP0U_F8462KfxQCuO2bHPGRe73Y3wt2uheQmKXdjlcM2V7Jk4Gepv4_F9nolvedov7cvV097BoVyXln9JZJG2clIpA1oiMg3OgwHvybLwGRR43FlF5KfVgG9abQZNFVRv2RuFMzP9mjwj9RwzvFL_7A0p_RMFfTSNELA</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>FrostNet: Towards Quantization-Aware Network Architecture Search</title><source>arXiv.org</source><creator>Kim, Taehoon ; Yoo, YoungJoon ; Yang, Jihoon</creator><creatorcontrib>Kim, Taehoon ; Yoo, YoungJoon ; Yang, Jihoon</creatorcontrib><description>INT8 quantization has become one of the standard techniques for deploying convolutional neural networks (CNNs) on edge devices to reduce the memory and computational resource usages. By analyzing quantized performances of existing mobile-target network architectures, we can raise an issue regarding the importance of network architecture for optimal INT8 quantization. In this paper, we present a new network architecture search (NAS) procedure to find a network that guarantees both full-precision (FLOAT32) and quantized (INT8) performances. We first propose critical but straightforward optimization method which enables quantization-aware training (QAT) : floating-point statistic assisting (StatAssist) and stochastic gradient boosting (GradBoost). By integrating the gradient-based NAS with StatAssist and GradBoost, we discovered a quantization-efficient network building block, Frost bottleneck. Furthermore, we used Frost bottleneck as the building block for hardware-aware NAS to obtain quantization-efficient networks, FrostNets, which show improved quantization performances compared to other mobile-target networks while maintaining competitive FLOAT32 performance. Our FrostNets achieve higher recognition accuracy than existing CNNs with comparable latency when quantized, due to higher latency reduction rate (average 65%).</description><identifier>DOI: 10.48550/arxiv.2006.09679</identifier><language>eng</language><subject>Computer Science - Computer Vision and Pattern Recognition ; Computer Science - Learning ; Statistics - Machine Learning</subject><creationdate>2020-06</creationdate><rights>http://creativecommons.org/licenses/by/4.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,782,887</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2006.09679$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2006.09679$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Kim, Taehoon</creatorcontrib><creatorcontrib>Yoo, YoungJoon</creatorcontrib><creatorcontrib>Yang, Jihoon</creatorcontrib><title>FrostNet: Towards Quantization-Aware Network Architecture Search</title><description>INT8 quantization has become one of the standard techniques for deploying convolutional neural networks (CNNs) on edge devices to reduce the memory and computational resource usages. By analyzing quantized performances of existing mobile-target network architectures, we can raise an issue regarding the importance of network architecture for optimal INT8 quantization. In this paper, we present a new network architecture search (NAS) procedure to find a network that guarantees both full-precision (FLOAT32) and quantized (INT8) performances. We first propose critical but straightforward optimization method which enables quantization-aware training (QAT) : floating-point statistic assisting (StatAssist) and stochastic gradient boosting (GradBoost). By integrating the gradient-based NAS with StatAssist and GradBoost, we discovered a quantization-efficient network building block, Frost bottleneck. Furthermore, we used Frost bottleneck as the building block for hardware-aware NAS to obtain quantization-efficient networks, FrostNets, which show improved quantization performances compared to other mobile-target networks while maintaining competitive FLOAT32 performance. Our FrostNets achieve higher recognition accuracy than existing CNNs with comparable latency when quantized, due to higher latency reduction rate (average 65%).</description><subject>Computer Science - Computer Vision and Pattern Recognition</subject><subject>Computer Science - Learning</subject><subject>Statistics - Machine Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2020</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj8FOwzAQRH3hgAofwKn5gYR11jYOJ6KKAlIFQs092tobYQENclwKfD1u4TSaGWlGT4gLCZWyWsMlxa_wWdUApoLGXDWn4mYZxyk9crouunFP0U_F8462KfxQCuO2bHPGRe73Y3wt2uheQmKXdjlcM2V7Jk4Gepv4_F9nolvedov7cvV097BoVyXln9JZJG2clIpA1oiMg3OgwHvybLwGRR43FlF5KfVgG9abQZNFVRv2RuFMzP9mjwj9RwzvFL_7A0p_RMFfTSNELA</recordid><startdate>20200617</startdate><enddate>20200617</enddate><creator>Kim, Taehoon</creator><creator>Yoo, YoungJoon</creator><creator>Yang, Jihoon</creator><scope>AKY</scope><scope>EPD</scope><scope>GOX</scope></search><sort><creationdate>20200617</creationdate><title>FrostNet: Towards Quantization-Aware Network Architecture Search</title><author>Kim, Taehoon ; Yoo, YoungJoon ; Yang, Jihoon</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a679-c83a56c114a01233e3fcc040ddade6d504ad3b8334d115f89e5bf5a83426ed643</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2020</creationdate><topic>Computer Science - Computer Vision and Pattern Recognition</topic><topic>Computer Science - Learning</topic><topic>Statistics - Machine Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Kim, Taehoon</creatorcontrib><creatorcontrib>Yoo, YoungJoon</creatorcontrib><creatorcontrib>Yang, Jihoon</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv Statistics</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Kim, Taehoon</au><au>Yoo, YoungJoon</au><au>Yang, Jihoon</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>FrostNet: Towards Quantization-Aware Network Architecture Search</atitle><date>2020-06-17</date><risdate>2020</risdate><abstract>INT8 quantization has become one of the standard techniques for deploying convolutional neural networks (CNNs) on edge devices to reduce the memory and computational resource usages. By analyzing quantized performances of existing mobile-target network architectures, we can raise an issue regarding the importance of network architecture for optimal INT8 quantization. In this paper, we present a new network architecture search (NAS) procedure to find a network that guarantees both full-precision (FLOAT32) and quantized (INT8) performances. We first propose critical but straightforward optimization method which enables quantization-aware training (QAT) : floating-point statistic assisting (StatAssist) and stochastic gradient boosting (GradBoost). By integrating the gradient-based NAS with StatAssist and GradBoost, we discovered a quantization-efficient network building block, Frost bottleneck. Furthermore, we used Frost bottleneck as the building block for hardware-aware NAS to obtain quantization-efficient networks, FrostNets, which show improved quantization performances compared to other mobile-target networks while maintaining competitive FLOAT32 performance. Our FrostNets achieve higher recognition accuracy than existing CNNs with comparable latency when quantized, due to higher latency reduction rate (average 65%).</abstract><doi>10.48550/arxiv.2006.09679</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2006.09679
ispartof
issn
language eng
recordid cdi_arxiv_primary_2006_09679
source arXiv.org
subjects Computer Science - Computer Vision and Pattern Recognition
Computer Science - Learning
Statistics - Machine Learning
title FrostNet: Towards Quantization-Aware Network Architecture Search
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-06T01%3A04%3A58IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=FrostNet:%20Towards%20Quantization-Aware%20Network%20Architecture%20Search&rft.au=Kim,%20Taehoon&rft.date=2020-06-17&rft_id=info:doi/10.48550/arxiv.2006.09679&rft_dat=%3Carxiv_GOX%3E2006_09679%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true