HyperSparse Neural Networks: Shifting Exploration to Exploitation through Adaptive Regularization

Sparse neural networks are a key factor in developing resource-efficient machine learning applications. We propose the novel and powerful sparse learning method Adaptive Regularized Training (ART) to compress dense into sparse networks. Instead of the commonly used binary mask during training to red...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Glandorf, Patrick, Kaiser, Timo, Rosenhahn, Bodo
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 Glandorf, Patrick
Kaiser, Timo
Rosenhahn, Bodo
description Sparse neural networks are a key factor in developing resource-efficient machine learning applications. We propose the novel and powerful sparse learning method Adaptive Regularized Training (ART) to compress dense into sparse networks. Instead of the commonly used binary mask during training to reduce the number of model weights, we inherently shrink weights close to zero in an iterative manner with increasing weight regularization. Our method compresses the pre-trained model knowledge into the weights of highest magnitude. Therefore, we introduce a novel regularization loss named HyperSparse that exploits the highest weights while conserving the ability of weight exploration. Extensive experiments on CIFAR and TinyImageNet show that our method leads to notable performance gains compared to other sparsification methods, especially in extremely high sparsity regimes up to 99.8 percent model sparsity. Additional investigations provide new insights into the patterns that are encoded in weights with high magnitudes.
doi_str_mv 10.48550/arxiv.2308.07163
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2308_07163</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2308_07163</sourcerecordid><originalsourceid>FETCH-LOGICAL-a673-82170768318f001f729ced5624b8c44cb19bbd7b1453dcd2810fa348992a442c3</originalsourceid><addsrcrecordid>eNotj99KwzAcRnPjhWw-gFfmBVrzr026uzGmE8YEt_vya5O0wbqUNJ2bT692uzocPvjgIPRISSpUlpFnCGd3ShknKiWS5vwewebSm7DvIQwG78wYoPtD_Pbhc1jgfetsdMcGr8995wNE5484-qu6ePM2-LFp8VJDH93J4A_TjB0E9zPtc3RnoRvMw40zdHhZH1abZPv--rZabhPIJU8Uo5LIXHGqLCHUSlbURmc5E5WqhagrWlSVlhUVGde1ZooSC1yoomAgBKv5DD1db6fGsg_uC8Kl_G8tp1b-C-uYUGc</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>HyperSparse Neural Networks: Shifting Exploration to Exploitation through Adaptive Regularization</title><source>arXiv.org</source><creator>Glandorf, Patrick ; Kaiser, Timo ; Rosenhahn, Bodo</creator><creatorcontrib>Glandorf, Patrick ; Kaiser, Timo ; Rosenhahn, Bodo</creatorcontrib><description>Sparse neural networks are a key factor in developing resource-efficient machine learning applications. We propose the novel and powerful sparse learning method Adaptive Regularized Training (ART) to compress dense into sparse networks. Instead of the commonly used binary mask during training to reduce the number of model weights, we inherently shrink weights close to zero in an iterative manner with increasing weight regularization. Our method compresses the pre-trained model knowledge into the weights of highest magnitude. Therefore, we introduce a novel regularization loss named HyperSparse that exploits the highest weights while conserving the ability of weight exploration. Extensive experiments on CIFAR and TinyImageNet show that our method leads to notable performance gains compared to other sparsification methods, especially in extremely high sparsity regimes up to 99.8 percent model sparsity. Additional investigations provide new insights into the patterns that are encoded in weights with high magnitudes.</description><identifier>DOI: 10.48550/arxiv.2308.07163</identifier><language>eng</language><subject>Computer Science - Computer Vision and Pattern Recognition</subject><creationdate>2023-08</creationdate><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,780,885</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2308.07163$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2308.07163$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Glandorf, Patrick</creatorcontrib><creatorcontrib>Kaiser, Timo</creatorcontrib><creatorcontrib>Rosenhahn, Bodo</creatorcontrib><title>HyperSparse Neural Networks: Shifting Exploration to Exploitation through Adaptive Regularization</title><description>Sparse neural networks are a key factor in developing resource-efficient machine learning applications. We propose the novel and powerful sparse learning method Adaptive Regularized Training (ART) to compress dense into sparse networks. Instead of the commonly used binary mask during training to reduce the number of model weights, we inherently shrink weights close to zero in an iterative manner with increasing weight regularization. Our method compresses the pre-trained model knowledge into the weights of highest magnitude. Therefore, we introduce a novel regularization loss named HyperSparse that exploits the highest weights while conserving the ability of weight exploration. Extensive experiments on CIFAR and TinyImageNet show that our method leads to notable performance gains compared to other sparsification methods, especially in extremely high sparsity regimes up to 99.8 percent model sparsity. Additional investigations provide new insights into the patterns that are encoded in weights with high magnitudes.</description><subject>Computer Science - Computer Vision and Pattern Recognition</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj99KwzAcRnPjhWw-gFfmBVrzr026uzGmE8YEt_vya5O0wbqUNJ2bT692uzocPvjgIPRISSpUlpFnCGd3ShknKiWS5vwewebSm7DvIQwG78wYoPtD_Pbhc1jgfetsdMcGr8995wNE5484-qu6ePM2-LFp8VJDH93J4A_TjB0E9zPtc3RnoRvMw40zdHhZH1abZPv--rZabhPIJU8Uo5LIXHGqLCHUSlbURmc5E5WqhagrWlSVlhUVGde1ZooSC1yoomAgBKv5DD1db6fGsg_uC8Kl_G8tp1b-C-uYUGc</recordid><startdate>20230814</startdate><enddate>20230814</enddate><creator>Glandorf, Patrick</creator><creator>Kaiser, Timo</creator><creator>Rosenhahn, Bodo</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20230814</creationdate><title>HyperSparse Neural Networks: Shifting Exploration to Exploitation through Adaptive Regularization</title><author>Glandorf, Patrick ; Kaiser, Timo ; Rosenhahn, Bodo</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a673-82170768318f001f729ced5624b8c44cb19bbd7b1453dcd2810fa348992a442c3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Computer Science - Computer Vision and Pattern Recognition</topic><toplevel>online_resources</toplevel><creatorcontrib>Glandorf, Patrick</creatorcontrib><creatorcontrib>Kaiser, Timo</creatorcontrib><creatorcontrib>Rosenhahn, Bodo</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Glandorf, Patrick</au><au>Kaiser, Timo</au><au>Rosenhahn, Bodo</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>HyperSparse Neural Networks: Shifting Exploration to Exploitation through Adaptive Regularization</atitle><date>2023-08-14</date><risdate>2023</risdate><abstract>Sparse neural networks are a key factor in developing resource-efficient machine learning applications. We propose the novel and powerful sparse learning method Adaptive Regularized Training (ART) to compress dense into sparse networks. Instead of the commonly used binary mask during training to reduce the number of model weights, we inherently shrink weights close to zero in an iterative manner with increasing weight regularization. Our method compresses the pre-trained model knowledge into the weights of highest magnitude. Therefore, we introduce a novel regularization loss named HyperSparse that exploits the highest weights while conserving the ability of weight exploration. Extensive experiments on CIFAR and TinyImageNet show that our method leads to notable performance gains compared to other sparsification methods, especially in extremely high sparsity regimes up to 99.8 percent model sparsity. Additional investigations provide new insights into the patterns that are encoded in weights with high magnitudes.</abstract><doi>10.48550/arxiv.2308.07163</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2308.07163
ispartof
issn
language eng
recordid cdi_arxiv_primary_2308_07163
source arXiv.org
subjects Computer Science - Computer Vision and Pattern Recognition
title HyperSparse Neural Networks: Shifting Exploration to Exploitation through Adaptive Regularization
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-13T11%3A18%3A08IST&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=HyperSparse%20Neural%20Networks:%20Shifting%20Exploration%20to%20Exploitation%20through%20Adaptive%20Regularization&rft.au=Glandorf,%20Patrick&rft.date=2023-08-14&rft_id=info:doi/10.48550/arxiv.2308.07163&rft_dat=%3Carxiv_GOX%3E2308_07163%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