Loading…

Designing Interpretable Recurrent Neural Networks for Video Reconstruction via Deep Unfolding

Deep unfolding methods design deep neural networks as learned variations of optimization algorithms through the unrolling of their iterations. These networks have been shown to achieve faster convergence and higher accuracy than the original optimization methods. In this line of research, this paper...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on image processing 2021, Vol.30, p.4099-4113
Main Authors: Luong, Huynh Van, Joukovsky, Boris, Deligiannis, Nikos
Format: Article
Language:English
Subjects:
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
cited_by cdi_FETCH-LOGICAL-c389t-d6d4d27fc6b2f6c40fe4fd0c4b6e3866b200fb4e5bad7dddce206cb9bfea65e23
cites cdi_FETCH-LOGICAL-c389t-d6d4d27fc6b2f6c40fe4fd0c4b6e3866b200fb4e5bad7dddce206cb9bfea65e23
container_end_page 4113
container_issue
container_start_page 4099
container_title IEEE transactions on image processing
container_volume 30
creator Luong, Huynh Van
Joukovsky, Boris
Deligiannis, Nikos
description Deep unfolding methods design deep neural networks as learned variations of optimization algorithms through the unrolling of their iterations. These networks have been shown to achieve faster convergence and higher accuracy than the original optimization methods. In this line of research, this paper presents novel interpretable deep recurrent neural networks (RNNs), designed by the unfolding of iterative algorithms that solve the task of sequential signal reconstruction (in particular, video reconstruction). The proposed networks are designed by accounting that video frames' patches have a sparse representation and the temporal difference between consecutive representations is also sparse. Specifically, we design an interpretable deep RNN (coined reweighted-RNN) by unrolling the iterations of a proximal method that solves a reweighted version of the \ell _{1} - \ell _{1} minimization problem. Due to the underlying minimization model, our reweighted-RNN has a different thresholding function (alias, different activation function) for each hidden unit in each layer. In this way, it has higher network expressivity than existing deep unfolding RNN models. We also present the derivative \ell _{1} - \ell _{1} -RNN model, which is obtained by unfolding a proximal method for the \ell _{1} - \ell _{1} minimization problem. We apply the proposed interpretable RNNs to the task of video frame reconstruction from low-dimensional measurements, that is, sequential video frame reconstruction. The experimental results on various datasets demonstrate that the proposed deep RNNs outperform various RNN models.
doi_str_mv 10.1109/TIP.2021.3069296
format article
fullrecord <record><control><sourceid>proquest_pubme</sourceid><recordid>TN_cdi_pubmed_primary_33798083</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><ieee_id>9394770</ieee_id><sourcerecordid>2512222171</sourcerecordid><originalsourceid>FETCH-LOGICAL-c389t-d6d4d27fc6b2f6c40fe4fd0c4b6e3866b200fb4e5bad7dddce206cb9bfea65e23</originalsourceid><addsrcrecordid>eNpdkE1LJDEQhoMoft8XFqTBi5eerXx0unOUUXcHBhUZvUnTnVSktScZk27Ff78ZZtbD1qWK5KmX4iHkB4UJpaB-LWb3EwaMTjhIxZTcIYdUCZoDCLabZijKvKRCHZCjGF8BqCio3CcHnJeqgoofkucrjN2L69xLNnMDhlXAoWl7zB5QjyGgG7JbHEPTpzZ8-vAWM-tD9tQZ9GvGuziEUQ-dd9lH12RXiKvs0Vnfm5R5QvZs00c83fZj8nhzvZj-yed3v2fTy3mueaWG3EgjDCutli2zUguwKKwBLVqJvJLpFcC2Aou2MaUxRiMDqVvVWmxkgYwfk4tN7ir49xHjUC-7qLHvG4d-jDUroCoqYJVK6Pl_6Ksfg0vXJYqyVLSkiYINpYOPMaCtV6FbNuGrplCv1ddJfb1WX2_Vp5WzbfDYLtF8L_xznYCfG6BDxO9vxZUoS-B_AfYViV4</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2512222171</pqid></control><display><type>article</type><title>Designing Interpretable Recurrent Neural Networks for Video Reconstruction via Deep Unfolding</title><source>IEEE Xplore (Online service)</source><creator>Luong, Huynh Van ; Joukovsky, Boris ; Deligiannis, Nikos</creator><creatorcontrib>Luong, Huynh Van ; Joukovsky, Boris ; Deligiannis, Nikos</creatorcontrib><description><![CDATA[Deep unfolding methods design deep neural networks as learned variations of optimization algorithms through the unrolling of their iterations. These networks have been shown to achieve faster convergence and higher accuracy than the original optimization methods. In this line of research, this paper presents novel interpretable deep recurrent neural networks (RNNs), designed by the unfolding of iterative algorithms that solve the task of sequential signal reconstruction (in particular, video reconstruction). The proposed networks are designed by accounting that video frames' patches have a sparse representation and the temporal difference between consecutive representations is also sparse. Specifically, we design an interpretable deep RNN (coined reweighted-RNN) by unrolling the iterations of a proximal method that solves a reweighted version of the <inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula>-<inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula> minimization problem. Due to the underlying minimization model, our reweighted-RNN has a different thresholding function (alias, different activation function) for each hidden unit in each layer. In this way, it has higher network expressivity than existing deep unfolding RNN models. We also present the derivative <inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula>-<inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula>-RNN model, which is obtained by unfolding a proximal method for the <inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula>-<inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula> minimization problem. We apply the proposed interpretable RNNs to the task of video frame reconstruction from low-dimensional measurements, that is, sequential video frame reconstruction. The experimental results on various datasets demonstrate that the proposed deep RNNs outperform various RNN models.]]></description><identifier>ISSN: 1057-7149</identifier><identifier>EISSN: 1941-0042</identifier><identifier>DOI: 10.1109/TIP.2021.3069296</identifier><identifier>PMID: 33798083</identifier><identifier>CODEN: IIPRE4</identifier><language>eng</language><publisher>United States: IEEE</publisher><subject>Algorithms ; Artificial neural networks ; Computational modeling ; Deep unfolding ; Design ; Dimensional measurement ; Image coding ; Image reconstruction ; Iterative algorithms ; Iterative methods ; Minimization ; Neural networks ; Optimization ; Recurrent neural networks ; Representations ; reweighted &lt;italic xmlns:ali="http://www.niso.org/schemas/ali/1.0/" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"&gt;ℓ ₁-&lt;italic xmlns:ali="http://www.niso.org/schemas/ali/1.0/" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"&gt;ℓ ₁ minimization ; sequential frame reconstruction ; Signal reconstruction ; Task analysis</subject><ispartof>IEEE transactions on image processing, 2021, Vol.30, p.4099-4113</ispartof><rights>Copyright The Institute of Electrical and Electronics Engineers, Inc. (IEEE) 2021</rights><lds50>peer_reviewed</lds50><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed><citedby>FETCH-LOGICAL-c389t-d6d4d27fc6b2f6c40fe4fd0c4b6e3866b200fb4e5bad7dddce206cb9bfea65e23</citedby><cites>FETCH-LOGICAL-c389t-d6d4d27fc6b2f6c40fe4fd0c4b6e3866b200fb4e5bad7dddce206cb9bfea65e23</cites><orcidid>0000-0001-9300-5860 ; 0000-0002-4562-9406</orcidid></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktohtml>$$Uhttps://ieeexplore.ieee.org/document/9394770$$EHTML$$P50$$Gieee$$H</linktohtml><link.rule.ids>314,780,784,4024,27923,27924,27925,54796</link.rule.ids><backlink>$$Uhttps://www.ncbi.nlm.nih.gov/pubmed/33798083$$D View this record in MEDLINE/PubMed$$Hfree_for_read</backlink></links><search><creatorcontrib>Luong, Huynh Van</creatorcontrib><creatorcontrib>Joukovsky, Boris</creatorcontrib><creatorcontrib>Deligiannis, Nikos</creatorcontrib><title>Designing Interpretable Recurrent Neural Networks for Video Reconstruction via Deep Unfolding</title><title>IEEE transactions on image processing</title><addtitle>TIP</addtitle><addtitle>IEEE Trans Image Process</addtitle><description><![CDATA[Deep unfolding methods design deep neural networks as learned variations of optimization algorithms through the unrolling of their iterations. These networks have been shown to achieve faster convergence and higher accuracy than the original optimization methods. In this line of research, this paper presents novel interpretable deep recurrent neural networks (RNNs), designed by the unfolding of iterative algorithms that solve the task of sequential signal reconstruction (in particular, video reconstruction). The proposed networks are designed by accounting that video frames' patches have a sparse representation and the temporal difference between consecutive representations is also sparse. Specifically, we design an interpretable deep RNN (coined reweighted-RNN) by unrolling the iterations of a proximal method that solves a reweighted version of the <inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula>-<inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula> minimization problem. Due to the underlying minimization model, our reweighted-RNN has a different thresholding function (alias, different activation function) for each hidden unit in each layer. In this way, it has higher network expressivity than existing deep unfolding RNN models. We also present the derivative <inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula>-<inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula>-RNN model, which is obtained by unfolding a proximal method for the <inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula>-<inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula> minimization problem. We apply the proposed interpretable RNNs to the task of video frame reconstruction from low-dimensional measurements, that is, sequential video frame reconstruction. The experimental results on various datasets demonstrate that the proposed deep RNNs outperform various RNN models.]]></description><subject>Algorithms</subject><subject>Artificial neural networks</subject><subject>Computational modeling</subject><subject>Deep unfolding</subject><subject>Design</subject><subject>Dimensional measurement</subject><subject>Image coding</subject><subject>Image reconstruction</subject><subject>Iterative algorithms</subject><subject>Iterative methods</subject><subject>Minimization</subject><subject>Neural networks</subject><subject>Optimization</subject><subject>Recurrent neural networks</subject><subject>Representations</subject><subject>reweighted &lt;italic xmlns:ali="http://www.niso.org/schemas/ali/1.0/" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"&gt;ℓ ₁-&lt;italic xmlns:ali="http://www.niso.org/schemas/ali/1.0/" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"&gt;ℓ ₁ minimization</subject><subject>sequential frame reconstruction</subject><subject>Signal reconstruction</subject><subject>Task analysis</subject><issn>1057-7149</issn><issn>1941-0042</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2021</creationdate><recordtype>article</recordtype><recordid>eNpdkE1LJDEQhoMoft8XFqTBi5eerXx0unOUUXcHBhUZvUnTnVSktScZk27Ff78ZZtbD1qWK5KmX4iHkB4UJpaB-LWb3EwaMTjhIxZTcIYdUCZoDCLabZijKvKRCHZCjGF8BqCio3CcHnJeqgoofkucrjN2L69xLNnMDhlXAoWl7zB5QjyGgG7JbHEPTpzZ8-vAWM-tD9tQZ9GvGuziEUQ-dd9lH12RXiKvs0Vnfm5R5QvZs00c83fZj8nhzvZj-yed3v2fTy3mueaWG3EgjDCutli2zUguwKKwBLVqJvJLpFcC2Aou2MaUxRiMDqVvVWmxkgYwfk4tN7ir49xHjUC-7qLHvG4d-jDUroCoqYJVK6Pl_6Ksfg0vXJYqyVLSkiYINpYOPMaCtV6FbNuGrplCv1ddJfb1WX2_Vp5WzbfDYLtF8L_xznYCfG6BDxO9vxZUoS-B_AfYViV4</recordid><startdate>2021</startdate><enddate>2021</enddate><creator>Luong, Huynh Van</creator><creator>Joukovsky, Boris</creator><creator>Deligiannis, Nikos</creator><general>IEEE</general><general>The Institute of Electrical and Electronics Engineers, Inc. (IEEE)</general><scope>97E</scope><scope>RIA</scope><scope>RIE</scope><scope>NPM</scope><scope>AAYXX</scope><scope>CITATION</scope><scope>7SC</scope><scope>7SP</scope><scope>8FD</scope><scope>JQ2</scope><scope>L7M</scope><scope>L~C</scope><scope>L~D</scope><scope>7X8</scope><orcidid>https://orcid.org/0000-0001-9300-5860</orcidid><orcidid>https://orcid.org/0000-0002-4562-9406</orcidid></search><sort><creationdate>2021</creationdate><title>Designing Interpretable Recurrent Neural Networks for Video Reconstruction via Deep Unfolding</title><author>Luong, Huynh Van ; Joukovsky, Boris ; Deligiannis, Nikos</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-c389t-d6d4d27fc6b2f6c40fe4fd0c4b6e3866b200fb4e5bad7dddce206cb9bfea65e23</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2021</creationdate><topic>Algorithms</topic><topic>Artificial neural networks</topic><topic>Computational modeling</topic><topic>Deep unfolding</topic><topic>Design</topic><topic>Dimensional measurement</topic><topic>Image coding</topic><topic>Image reconstruction</topic><topic>Iterative algorithms</topic><topic>Iterative methods</topic><topic>Minimization</topic><topic>Neural networks</topic><topic>Optimization</topic><topic>Recurrent neural networks</topic><topic>Representations</topic><topic>reweighted &lt;italic xmlns:ali="http://www.niso.org/schemas/ali/1.0/" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"&gt;ℓ ₁-&lt;italic xmlns:ali="http://www.niso.org/schemas/ali/1.0/" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"&gt;ℓ ₁ minimization</topic><topic>sequential frame reconstruction</topic><topic>Signal reconstruction</topic><topic>Task analysis</topic><toplevel>peer_reviewed</toplevel><toplevel>online_resources</toplevel><creatorcontrib>Luong, Huynh Van</creatorcontrib><creatorcontrib>Joukovsky, Boris</creatorcontrib><creatorcontrib>Deligiannis, Nikos</creatorcontrib><collection>IEEE All-Society Periodicals Package (ASPP) 2005-present</collection><collection>IEEE All-Society Periodicals Package (ASPP) 1998-Present</collection><collection>IEEE Electronic Library Online</collection><collection>PubMed</collection><collection>CrossRef</collection><collection>Computer and Information Systems Abstracts</collection><collection>Electronics &amp; Communications Abstracts</collection><collection>Technology Research Database</collection><collection>ProQuest Computer Science Collection</collection><collection>Advanced Technologies Database with Aerospace</collection><collection>Computer and Information Systems Abstracts – Academic</collection><collection>Computer and Information Systems Abstracts Professional</collection><collection>MEDLINE - Academic</collection><jtitle>IEEE transactions on image processing</jtitle></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Luong, Huynh Van</au><au>Joukovsky, Boris</au><au>Deligiannis, Nikos</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Designing Interpretable Recurrent Neural Networks for Video Reconstruction via Deep Unfolding</atitle><jtitle>IEEE transactions on image processing</jtitle><stitle>TIP</stitle><addtitle>IEEE Trans Image Process</addtitle><date>2021</date><risdate>2021</risdate><volume>30</volume><spage>4099</spage><epage>4113</epage><pages>4099-4113</pages><issn>1057-7149</issn><eissn>1941-0042</eissn><coden>IIPRE4</coden><abstract><![CDATA[Deep unfolding methods design deep neural networks as learned variations of optimization algorithms through the unrolling of their iterations. These networks have been shown to achieve faster convergence and higher accuracy than the original optimization methods. In this line of research, this paper presents novel interpretable deep recurrent neural networks (RNNs), designed by the unfolding of iterative algorithms that solve the task of sequential signal reconstruction (in particular, video reconstruction). The proposed networks are designed by accounting that video frames' patches have a sparse representation and the temporal difference between consecutive representations is also sparse. Specifically, we design an interpretable deep RNN (coined reweighted-RNN) by unrolling the iterations of a proximal method that solves a reweighted version of the <inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula>-<inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula> minimization problem. Due to the underlying minimization model, our reweighted-RNN has a different thresholding function (alias, different activation function) for each hidden unit in each layer. In this way, it has higher network expressivity than existing deep unfolding RNN models. We also present the derivative <inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula>-<inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula>-RNN model, which is obtained by unfolding a proximal method for the <inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula>-<inline-formula> <tex-math notation="LaTeX">\ell _{1} </tex-math></inline-formula> minimization problem. We apply the proposed interpretable RNNs to the task of video frame reconstruction from low-dimensional measurements, that is, sequential video frame reconstruction. The experimental results on various datasets demonstrate that the proposed deep RNNs outperform various RNN models.]]></abstract><cop>United States</cop><pub>IEEE</pub><pmid>33798083</pmid><doi>10.1109/TIP.2021.3069296</doi><tpages>15</tpages><orcidid>https://orcid.org/0000-0001-9300-5860</orcidid><orcidid>https://orcid.org/0000-0002-4562-9406</orcidid><oa>free_for_read</oa></addata></record>
fulltext fulltext
identifier ISSN: 1057-7149
ispartof IEEE transactions on image processing, 2021, Vol.30, p.4099-4113
issn 1057-7149
1941-0042
language eng
recordid cdi_pubmed_primary_33798083
source IEEE Xplore (Online service)
subjects Algorithms
Artificial neural networks
Computational modeling
Deep unfolding
Design
Dimensional measurement
Image coding
Image reconstruction
Iterative algorithms
Iterative methods
Minimization
Neural networks
Optimization
Recurrent neural networks
Representations
reweighted <italic xmlns:ali="http://www.niso.org/schemas/ali/1.0/" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">ℓ ₁-<italic xmlns:ali="http://www.niso.org/schemas/ali/1.0/" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">ℓ ₁ minimization
sequential frame reconstruction
Signal reconstruction
Task analysis
title Designing Interpretable Recurrent Neural Networks for Video Reconstruction via Deep Unfolding
url http://sfxeu10.hosted.exlibrisgroup.com/loughborough?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-25T19%3A33%3A02IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest_pubme&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Designing%20Interpretable%20Recurrent%20Neural%20Networks%20for%20Video%20Reconstruction%20via%20Deep%20Unfolding&rft.jtitle=IEEE%20transactions%20on%20image%20processing&rft.au=Luong,%20Huynh%20Van&rft.date=2021&rft.volume=30&rft.spage=4099&rft.epage=4113&rft.pages=4099-4113&rft.issn=1057-7149&rft.eissn=1941-0042&rft.coden=IIPRE4&rft_id=info:doi/10.1109/TIP.2021.3069296&rft_dat=%3Cproquest_pubme%3E2512222171%3C/proquest_pubme%3E%3Cgrp_id%3Ecdi_FETCH-LOGICAL-c389t-d6d4d27fc6b2f6c40fe4fd0c4b6e3866b200fb4e5bad7dddce206cb9bfea65e23%3C/grp_id%3E%3Coa%3E%3C/oa%3E%3Curl%3E%3C/url%3E&rft_id=info:oai/&rft_pqid=2512222171&rft_id=info:pmid/33798083&rft_ieee_id=9394770&rfr_iscdi=true