Dueling Decoders: Regularizing Variational Autoencoder Latent Spaces

Variational autoencoders learn unsupervised data representations, but these models frequently converge to minima that fail to preserve meaningful semantic information. For example, variational autoencoders with autoregressive decoders often collapse into autodecoders, where they learn to ignore the...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Seybold, Bryan, Fertig, Emily, Alemi, Alex, Fischer, Ian
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 Seybold, Bryan
Fertig, Emily
Alemi, Alex
Fischer, Ian
description Variational autoencoders learn unsupervised data representations, but these models frequently converge to minima that fail to preserve meaningful semantic information. For example, variational autoencoders with autoregressive decoders often collapse into autodecoders, where they learn to ignore the encoder input. In this work, we demonstrate that adding an auxiliary decoder to regularize the latent space can prevent this collapse, but successful auxiliary decoding tasks are domain dependent. Auxiliary decoders can increase the amount of semantic information encoded in the latent space and visible in the reconstructions. The semantic information in the variational autoencoder's representation is only weakly correlated with its rate, distortion, or evidence lower bound. Compared to other popular strategies that modify the training objective, our regularization of the latent space generally increased the semantic information content.
doi_str_mv 10.48550/arxiv.1905.07478
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_1905_07478</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>1905_07478</sourcerecordid><originalsourceid>FETCH-LOGICAL-a678-9f451889b0b1f698113c6c1e90025269100d8b3783ae6beb41f11ff49a0b8ecc3</originalsourceid><addsrcrecordid>eNotj81KxDAUhbNxIaMP4Mq8QGvu5KeJu2HqHxQEHdyWm8zNEKjtkLaiPr1OdXUOh48DH2NXIEpltRY3mD_TRwlO6FJUqrLnrK5n6lJ_4DWFYU95vOUvdJg7zOn7NL_9FpzS0GPHN_M0UL9gvMGJ-om_HjHQeMHOInYjXf7niu3u73bbx6J5fnjabpoCTWULF5UGa50XHqJxFkAGE4CcEGu9Ng6E2FsvKyuRjCevIALEqBwKbykEuWLXf7eLRnvM6R3zV3vSaRcd-QNKpkUY</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Dueling Decoders: Regularizing Variational Autoencoder Latent Spaces</title><source>arXiv.org</source><creator>Seybold, Bryan ; Fertig, Emily ; Alemi, Alex ; Fischer, Ian</creator><creatorcontrib>Seybold, Bryan ; Fertig, Emily ; Alemi, Alex ; Fischer, Ian</creatorcontrib><description>Variational autoencoders learn unsupervised data representations, but these models frequently converge to minima that fail to preserve meaningful semantic information. For example, variational autoencoders with autoregressive decoders often collapse into autodecoders, where they learn to ignore the encoder input. In this work, we demonstrate that adding an auxiliary decoder to regularize the latent space can prevent this collapse, but successful auxiliary decoding tasks are domain dependent. Auxiliary decoders can increase the amount of semantic information encoded in the latent space and visible in the reconstructions. The semantic information in the variational autoencoder's representation is only weakly correlated with its rate, distortion, or evidence lower bound. Compared to other popular strategies that modify the training objective, our regularization of the latent space generally increased the semantic information content.</description><identifier>DOI: 10.48550/arxiv.1905.07478</identifier><language>eng</language><subject>Computer Science - Learning ; Statistics - Machine Learning</subject><creationdate>2019-05</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/1905.07478$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.1905.07478$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Seybold, Bryan</creatorcontrib><creatorcontrib>Fertig, Emily</creatorcontrib><creatorcontrib>Alemi, Alex</creatorcontrib><creatorcontrib>Fischer, Ian</creatorcontrib><title>Dueling Decoders: Regularizing Variational Autoencoder Latent Spaces</title><description>Variational autoencoders learn unsupervised data representations, but these models frequently converge to minima that fail to preserve meaningful semantic information. For example, variational autoencoders with autoregressive decoders often collapse into autodecoders, where they learn to ignore the encoder input. In this work, we demonstrate that adding an auxiliary decoder to regularize the latent space can prevent this collapse, but successful auxiliary decoding tasks are domain dependent. Auxiliary decoders can increase the amount of semantic information encoded in the latent space and visible in the reconstructions. The semantic information in the variational autoencoder's representation is only weakly correlated with its rate, distortion, or evidence lower bound. Compared to other popular strategies that modify the training objective, our regularization of the latent space generally increased the semantic information content.</description><subject>Computer Science - Learning</subject><subject>Statistics - Machine Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2019</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj81KxDAUhbNxIaMP4Mq8QGvu5KeJu2HqHxQEHdyWm8zNEKjtkLaiPr1OdXUOh48DH2NXIEpltRY3mD_TRwlO6FJUqrLnrK5n6lJ_4DWFYU95vOUvdJg7zOn7NL_9FpzS0GPHN_M0UL9gvMGJ-om_HjHQeMHOInYjXf7niu3u73bbx6J5fnjabpoCTWULF5UGa50XHqJxFkAGE4CcEGu9Ng6E2FsvKyuRjCevIALEqBwKbykEuWLXf7eLRnvM6R3zV3vSaRcd-QNKpkUY</recordid><startdate>20190517</startdate><enddate>20190517</enddate><creator>Seybold, Bryan</creator><creator>Fertig, Emily</creator><creator>Alemi, Alex</creator><creator>Fischer, Ian</creator><scope>AKY</scope><scope>EPD</scope><scope>GOX</scope></search><sort><creationdate>20190517</creationdate><title>Dueling Decoders: Regularizing Variational Autoencoder Latent Spaces</title><author>Seybold, Bryan ; Fertig, Emily ; Alemi, Alex ; Fischer, Ian</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a678-9f451889b0b1f698113c6c1e90025269100d8b3783ae6beb41f11ff49a0b8ecc3</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2019</creationdate><topic>Computer Science - Learning</topic><topic>Statistics - Machine Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Seybold, Bryan</creatorcontrib><creatorcontrib>Fertig, Emily</creatorcontrib><creatorcontrib>Alemi, Alex</creatorcontrib><creatorcontrib>Fischer, Ian</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>Seybold, Bryan</au><au>Fertig, Emily</au><au>Alemi, Alex</au><au>Fischer, Ian</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Dueling Decoders: Regularizing Variational Autoencoder Latent Spaces</atitle><date>2019-05-17</date><risdate>2019</risdate><abstract>Variational autoencoders learn unsupervised data representations, but these models frequently converge to minima that fail to preserve meaningful semantic information. For example, variational autoencoders with autoregressive decoders often collapse into autodecoders, where they learn to ignore the encoder input. In this work, we demonstrate that adding an auxiliary decoder to regularize the latent space can prevent this collapse, but successful auxiliary decoding tasks are domain dependent. Auxiliary decoders can increase the amount of semantic information encoded in the latent space and visible in the reconstructions. The semantic information in the variational autoencoder's representation is only weakly correlated with its rate, distortion, or evidence lower bound. Compared to other popular strategies that modify the training objective, our regularization of the latent space generally increased the semantic information content.</abstract><doi>10.48550/arxiv.1905.07478</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.1905.07478
ispartof
issn
language eng
recordid cdi_arxiv_primary_1905_07478
source arXiv.org
subjects Computer Science - Learning
Statistics - Machine Learning
title Dueling Decoders: Regularizing Variational Autoencoder Latent Spaces
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-02T06%3A09%3A46IST&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=Dueling%20Decoders:%20Regularizing%20Variational%20Autoencoder%20Latent%20Spaces&rft.au=Seybold,%20Bryan&rft.date=2019-05-17&rft_id=info:doi/10.48550/arxiv.1905.07478&rft_dat=%3Carxiv_GOX%3E1905_07478%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