Variational Potential Flow: A Novel Probabilistic Framework for Energy-Based Generative Modelling

Energy based models (EBMs) are appealing for their generality and simplicity in data likelihood modeling, but have conventionally been difficult to train due to the unstable and time-consuming implicit MCMC sampling during contrastive divergence training. In this paper, we present a novel energy-bas...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:arXiv.org 2024-07
Hauptverfasser: Junn Yong Loo, Adeline, Michelle, Pal, Arghya, Baskaran, Vishnu Monn, Chee-Ming, Ting, Phan, Raphael C -W
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title arXiv.org
container_volume
creator Junn Yong Loo
Adeline, Michelle
Pal, Arghya
Baskaran, Vishnu Monn
Chee-Ming, Ting
Phan, Raphael C -W
description Energy based models (EBMs) are appealing for their generality and simplicity in data likelihood modeling, but have conventionally been difficult to train due to the unstable and time-consuming implicit MCMC sampling during contrastive divergence training. In this paper, we present a novel energy-based generative framework, Variational Potential Flow (VAPO), that entirely dispenses with implicit MCMC sampling and does not rely on complementary latent models or cooperative training. The VAPO framework aims to learn a potential energy function whose gradient (flow) guides the prior samples, so that their density evolution closely follows an approximate data likelihood homotopy. An energy loss function is then formulated to minimize the Kullback-Leibler divergence between density evolution of the flow-driven prior and the data likelihood homotopy. Images can be generated after training the potential energy, by initializing the samples from Gaussian prior and solving the ODE governing the potential flow on a fixed time interval using generic ODE solvers. Experiment results show that the proposed VAPO framework is capable of generating realistic images on various image datasets. In particular, our proposed framework achieves competitive FID scores for unconditional image generation on the CIFAR-10 and CelebA datasets.
format Article
fullrecord <record><control><sourceid>proquest</sourceid><recordid>TN_cdi_proquest_journals_3083765155</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>3083765155</sourcerecordid><originalsourceid>FETCH-proquest_journals_30837651553</originalsourceid><addsrcrecordid>eNqNTksKwjAUDIJg0d4h4LrQJvaDO5VWN4oLcVtS-1pSY58macXbm4UHcDUfhpmZEI9xHgXZirEZ8Y3pwjBkScrimHtEXIWWwkrshaJntNBb6Vih8L2mG3rCEZyvsRKVVNJYeaOFFg94o77TBjXNe9DtJ9gKAzXdg1OubQR6xBqUkn27INNGKAP-D-dkWeSX3SF4anwNYGzZ4aDdvCl5mPE0iSP37L_UFxo_RRQ</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>3083765155</pqid></control><display><type>article</type><title>Variational Potential Flow: A Novel Probabilistic Framework for Energy-Based Generative Modelling</title><source>Free E- Journals</source><creator>Junn Yong Loo ; Adeline, Michelle ; Pal, Arghya ; Baskaran, Vishnu Monn ; Chee-Ming, Ting ; Phan, Raphael C -W</creator><creatorcontrib>Junn Yong Loo ; Adeline, Michelle ; Pal, Arghya ; Baskaran, Vishnu Monn ; Chee-Ming, Ting ; Phan, Raphael C -W</creatorcontrib><description>Energy based models (EBMs) are appealing for their generality and simplicity in data likelihood modeling, but have conventionally been difficult to train due to the unstable and time-consuming implicit MCMC sampling during contrastive divergence training. In this paper, we present a novel energy-based generative framework, Variational Potential Flow (VAPO), that entirely dispenses with implicit MCMC sampling and does not rely on complementary latent models or cooperative training. The VAPO framework aims to learn a potential energy function whose gradient (flow) guides the prior samples, so that their density evolution closely follows an approximate data likelihood homotopy. An energy loss function is then formulated to minimize the Kullback-Leibler divergence between density evolution of the flow-driven prior and the data likelihood homotopy. Images can be generated after training the potential energy, by initializing the samples from Gaussian prior and solving the ODE governing the potential flow on a fixed time interval using generic ODE solvers. Experiment results show that the proposed VAPO framework is capable of generating realistic images on various image datasets. In particular, our proposed framework achieves competitive FID scores for unconditional image generation on the CIFAR-10 and CelebA datasets.</description><identifier>EISSN: 2331-8422</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Datasets ; Density ; Energy ; Evolution ; Image processing ; Potential energy ; Potential flow ; Sampling</subject><ispartof>arXiv.org, 2024-07</ispartof><rights>2024. This work is published under http://creativecommons.org/licenses/by-nc-nd/4.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.</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>778,782</link.rule.ids></links><search><creatorcontrib>Junn Yong Loo</creatorcontrib><creatorcontrib>Adeline, Michelle</creatorcontrib><creatorcontrib>Pal, Arghya</creatorcontrib><creatorcontrib>Baskaran, Vishnu Monn</creatorcontrib><creatorcontrib>Chee-Ming, Ting</creatorcontrib><creatorcontrib>Phan, Raphael C -W</creatorcontrib><title>Variational Potential Flow: A Novel Probabilistic Framework for Energy-Based Generative Modelling</title><title>arXiv.org</title><description>Energy based models (EBMs) are appealing for their generality and simplicity in data likelihood modeling, but have conventionally been difficult to train due to the unstable and time-consuming implicit MCMC sampling during contrastive divergence training. In this paper, we present a novel energy-based generative framework, Variational Potential Flow (VAPO), that entirely dispenses with implicit MCMC sampling and does not rely on complementary latent models or cooperative training. The VAPO framework aims to learn a potential energy function whose gradient (flow) guides the prior samples, so that their density evolution closely follows an approximate data likelihood homotopy. An energy loss function is then formulated to minimize the Kullback-Leibler divergence between density evolution of the flow-driven prior and the data likelihood homotopy. Images can be generated after training the potential energy, by initializing the samples from Gaussian prior and solving the ODE governing the potential flow on a fixed time interval using generic ODE solvers. Experiment results show that the proposed VAPO framework is capable of generating realistic images on various image datasets. In particular, our proposed framework achieves competitive FID scores for unconditional image generation on the CIFAR-10 and CelebA datasets.</description><subject>Datasets</subject><subject>Density</subject><subject>Energy</subject><subject>Evolution</subject><subject>Image processing</subject><subject>Potential energy</subject><subject>Potential flow</subject><subject>Sampling</subject><issn>2331-8422</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>ABUWG</sourceid><sourceid>AFKRA</sourceid><sourceid>AZQEC</sourceid><sourceid>BENPR</sourceid><sourceid>CCPQU</sourceid><sourceid>DWQXO</sourceid><recordid>eNqNTksKwjAUDIJg0d4h4LrQJvaDO5VWN4oLcVtS-1pSY58macXbm4UHcDUfhpmZEI9xHgXZirEZ8Y3pwjBkScrimHtEXIWWwkrshaJntNBb6Vih8L2mG3rCEZyvsRKVVNJYeaOFFg94o77TBjXNe9DtJ9gKAzXdg1OubQR6xBqUkn27INNGKAP-D-dkWeSX3SF4anwNYGzZ4aDdvCl5mPE0iSP37L_UFxo_RRQ</recordid><startdate>20240721</startdate><enddate>20240721</enddate><creator>Junn Yong Loo</creator><creator>Adeline, Michelle</creator><creator>Pal, Arghya</creator><creator>Baskaran, Vishnu Monn</creator><creator>Chee-Ming, Ting</creator><creator>Phan, Raphael C -W</creator><general>Cornell University Library, arXiv.org</general><scope>8FE</scope><scope>8FG</scope><scope>ABJCF</scope><scope>ABUWG</scope><scope>AFKRA</scope><scope>AZQEC</scope><scope>BENPR</scope><scope>BGLVJ</scope><scope>CCPQU</scope><scope>DWQXO</scope><scope>HCIFZ</scope><scope>L6V</scope><scope>M7S</scope><scope>PIMPY</scope><scope>PQEST</scope><scope>PQQKQ</scope><scope>PQUKI</scope><scope>PRINS</scope><scope>PTHSS</scope></search><sort><creationdate>20240721</creationdate><title>Variational Potential Flow: A Novel Probabilistic Framework for Energy-Based Generative Modelling</title><author>Junn Yong Loo ; Adeline, Michelle ; Pal, Arghya ; Baskaran, Vishnu Monn ; Chee-Ming, Ting ; Phan, Raphael C -W</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-proquest_journals_30837651553</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Datasets</topic><topic>Density</topic><topic>Energy</topic><topic>Evolution</topic><topic>Image processing</topic><topic>Potential energy</topic><topic>Potential flow</topic><topic>Sampling</topic><toplevel>online_resources</toplevel><creatorcontrib>Junn Yong Loo</creatorcontrib><creatorcontrib>Adeline, Michelle</creatorcontrib><creatorcontrib>Pal, Arghya</creatorcontrib><creatorcontrib>Baskaran, Vishnu Monn</creatorcontrib><creatorcontrib>Chee-Ming, Ting</creatorcontrib><creatorcontrib>Phan, Raphael C -W</creatorcontrib><collection>ProQuest SciTech Collection</collection><collection>ProQuest Technology Collection</collection><collection>Materials Science &amp; Engineering Collection</collection><collection>ProQuest Central (Alumni Edition)</collection><collection>ProQuest Central UK/Ireland</collection><collection>ProQuest Central Essentials</collection><collection>ProQuest Central</collection><collection>Technology Collection</collection><collection>ProQuest One Community College</collection><collection>ProQuest Central Korea</collection><collection>SciTech Premium Collection</collection><collection>ProQuest Engineering Collection</collection><collection>Engineering Database</collection><collection>Publicly Available Content Database</collection><collection>ProQuest One Academic Eastern Edition (DO NOT USE)</collection><collection>ProQuest One Academic</collection><collection>ProQuest One Academic UKI Edition</collection><collection>ProQuest Central China</collection><collection>Engineering Collection</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Junn Yong Loo</au><au>Adeline, Michelle</au><au>Pal, Arghya</au><au>Baskaran, Vishnu Monn</au><au>Chee-Ming, Ting</au><au>Phan, Raphael C -W</au><format>book</format><genre>document</genre><ristype>GEN</ristype><atitle>Variational Potential Flow: A Novel Probabilistic Framework for Energy-Based Generative Modelling</atitle><jtitle>arXiv.org</jtitle><date>2024-07-21</date><risdate>2024</risdate><eissn>2331-8422</eissn><abstract>Energy based models (EBMs) are appealing for their generality and simplicity in data likelihood modeling, but have conventionally been difficult to train due to the unstable and time-consuming implicit MCMC sampling during contrastive divergence training. In this paper, we present a novel energy-based generative framework, Variational Potential Flow (VAPO), that entirely dispenses with implicit MCMC sampling and does not rely on complementary latent models or cooperative training. The VAPO framework aims to learn a potential energy function whose gradient (flow) guides the prior samples, so that their density evolution closely follows an approximate data likelihood homotopy. An energy loss function is then formulated to minimize the Kullback-Leibler divergence between density evolution of the flow-driven prior and the data likelihood homotopy. Images can be generated after training the potential energy, by initializing the samples from Gaussian prior and solving the ODE governing the potential flow on a fixed time interval using generic ODE solvers. Experiment results show that the proposed VAPO framework is capable of generating realistic images on various image datasets. In particular, our proposed framework achieves competitive FID scores for unconditional image generation on the CIFAR-10 and CelebA datasets.</abstract><cop>Ithaca</cop><pub>Cornell University Library, arXiv.org</pub><oa>free_for_read</oa></addata></record>
fulltext fulltext
identifier EISSN: 2331-8422
ispartof arXiv.org, 2024-07
issn 2331-8422
language eng
recordid cdi_proquest_journals_3083765155
source Free E- Journals
subjects Datasets
Density
Energy
Evolution
Image processing
Potential energy
Potential flow
Sampling
title Variational Potential Flow: A Novel Probabilistic Framework for Energy-Based Generative Modelling
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-16T21%3A20%3A15IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=document&rft.atitle=Variational%20Potential%20Flow:%20A%20Novel%20Probabilistic%20Framework%20for%20Energy-Based%20Generative%20Modelling&rft.jtitle=arXiv.org&rft.au=Junn%20Yong%20Loo&rft.date=2024-07-21&rft.eissn=2331-8422&rft_id=info:doi/&rft_dat=%3Cproquest%3E3083765155%3C/proquest%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=3083765155&rft_id=info:pmid/&rfr_iscdi=true