Loading…
The Bethe approximation of the pattern maximum likelihood distribution
Among all memoryless source distributions, the pattern maximum likelihood (PML) distribution is the distribution which maximizes the probability that a memoryless source produces a string with a given pattern. Equivalently, the PML distribution maximizes the permanent of a certain non-negative matri...
Saved in:
Main Author: | |
---|---|
Format: | Conference Proceeding |
Language: | English |
Subjects: | |
Online Access: | Request full text |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
cited_by | |
---|---|
cites | |
container_end_page | 2016 |
container_issue | |
container_start_page | 2012 |
container_title | |
container_volume | |
creator | Vontobel, P. O. |
description | Among all memoryless source distributions, the pattern maximum likelihood (PML) distribution is the distribution which maximizes the probability that a memoryless source produces a string with a given pattern. Equivalently, the PML distribution maximizes the permanent of a certain non-negative matrix. We reformulate this maximization problem as a double minimization problem of a suitable Gibbs free energy function. Because finding the minimum of this function appears intractable for practically relevant problem sizes, one must look for tractable approximations. One approach is to approximately find a minimum (or at least a local minimum) of the Gibbs free energy function by applying an alternating minimization algorithm where the steps are based on quantities that are obtained by Markov chain Monte Carlo sampling. One can show that this approach is equivalent to an algorithm that was proposed by Orlitsky et al. An alternative approach is to replace the Gibbs free energy function by a tractable approximation like the Bethe free energy function and to apply an alternating minimization algorithm to this function. As it turns out, empirically, this latter approach gives very good approximations to the PML distribution (or at least a locally optimal PML distribution), and, for the same level of accuracy, is two to three orders of magnitude faster than the former approach for practically relevant problem sizes. Moreover, the above free energy framework allows us to simplify some earlier proofs of properties of the PML distribution and to derive some new properties of the PML distribution, along with obtaining similar results for its Bethe approximation. |
doi_str_mv | 10.1109/ISIT.2012.6283654 |
format | conference_proceeding |
fullrecord | <record><control><sourceid>ieee_6IE</sourceid><recordid>TN_cdi_ieee_primary_6283654</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><ieee_id>6283654</ieee_id><sourcerecordid>6283654</sourcerecordid><originalsourceid>FETCH-LOGICAL-i175t-d8ba459e55f8c3d166646add6490870f0c5a1e8f0a3b83bdf64f38512ed62be53</originalsourceid><addsrcrecordid>eNpNkN1KxDAQheMfuKx9APEmL9Ca_6SXurhaWPDCer2kmwkbbbclzYK-vS2u4FycgfkOc4ZB6JaSglJS3ldvVV0wQlmhmOFKijOUldpQoTRnUht-jhaMSp0bSvXFf2aIuvxjpJTXKBvHDzKVnq1ygdb1HvAjpEntMMT-K3Q2hf6Ae4_n4WBTgnjAnZ3IscNt-IQ27PveYRfGFENznO036MrbdoTs1Jfoff1Ur17yzetztXrY5GFKS7kzjRWyBCm92XFHlVJCWeeUKInRxJOdtBSMJ5Y3hjfOK-G5kZSBU6wByZfo7ndvAIDtEKdr4_f29BX-A3acUfU</addsrcrecordid><sourcetype>Publisher</sourcetype><iscdi>true</iscdi><recordtype>conference_proceeding</recordtype></control><display><type>conference_proceeding</type><title>The Bethe approximation of the pattern maximum likelihood distribution</title><source>IEEE Electronic Library (IEL) Conference Proceedings</source><creator>Vontobel, P. O.</creator><creatorcontrib>Vontobel, P. O.</creatorcontrib><description>Among all memoryless source distributions, the pattern maximum likelihood (PML) distribution is the distribution which maximizes the probability that a memoryless source produces a string with a given pattern. Equivalently, the PML distribution maximizes the permanent of a certain non-negative matrix. We reformulate this maximization problem as a double minimization problem of a suitable Gibbs free energy function. Because finding the minimum of this function appears intractable for practically relevant problem sizes, one must look for tractable approximations. One approach is to approximately find a minimum (or at least a local minimum) of the Gibbs free energy function by applying an alternating minimization algorithm where the steps are based on quantities that are obtained by Markov chain Monte Carlo sampling. One can show that this approach is equivalent to an algorithm that was proposed by Orlitsky et al. An alternative approach is to replace the Gibbs free energy function by a tractable approximation like the Bethe free energy function and to apply an alternating minimization algorithm to this function. As it turns out, empirically, this latter approach gives very good approximations to the PML distribution (or at least a locally optimal PML distribution), and, for the same level of accuracy, is two to three orders of magnitude faster than the former approach for practically relevant problem sizes. Moreover, the above free energy framework allows us to simplify some earlier proofs of properties of the PML distribution and to derive some new properties of the PML distribution, along with obtaining similar results for its Bethe approximation.</description><identifier>ISSN: 2157-8095</identifier><identifier>ISBN: 9781467325806</identifier><identifier>ISBN: 1467325805</identifier><identifier>EISSN: 2157-8117</identifier><identifier>EISBN: 9781467325783</identifier><identifier>EISBN: 1467325791</identifier><identifier>EISBN: 9781467325790</identifier><identifier>EISBN: 1467325783</identifier><identifier>DOI: 10.1109/ISIT.2012.6283654</identifier><language>eng</language><publisher>IEEE</publisher><subject>Approximation algorithms ; Approximation methods ; Graphical models ; Maximum likelihood estimation ; Mercury (metals) ; Minimization ; Vectors</subject><ispartof>2012 IEEE International Symposium on Information Theory Proceedings, 2012, p.2012-2016</ispartof><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><linktohtml>$$Uhttps://ieeexplore.ieee.org/document/6283654$$EHTML$$P50$$Gieee$$H</linktohtml><link.rule.ids>309,310,780,784,789,790,2058,27925,54555,54920,54932</link.rule.ids><linktorsrc>$$Uhttps://ieeexplore.ieee.org/document/6283654$$EView_record_in_IEEE$$FView_record_in_$$GIEEE</linktorsrc></links><search><creatorcontrib>Vontobel, P. O.</creatorcontrib><title>The Bethe approximation of the pattern maximum likelihood distribution</title><title>2012 IEEE International Symposium on Information Theory Proceedings</title><addtitle>ISIT</addtitle><description>Among all memoryless source distributions, the pattern maximum likelihood (PML) distribution is the distribution which maximizes the probability that a memoryless source produces a string with a given pattern. Equivalently, the PML distribution maximizes the permanent of a certain non-negative matrix. We reformulate this maximization problem as a double minimization problem of a suitable Gibbs free energy function. Because finding the minimum of this function appears intractable for practically relevant problem sizes, one must look for tractable approximations. One approach is to approximately find a minimum (or at least a local minimum) of the Gibbs free energy function by applying an alternating minimization algorithm where the steps are based on quantities that are obtained by Markov chain Monte Carlo sampling. One can show that this approach is equivalent to an algorithm that was proposed by Orlitsky et al. An alternative approach is to replace the Gibbs free energy function by a tractable approximation like the Bethe free energy function and to apply an alternating minimization algorithm to this function. As it turns out, empirically, this latter approach gives very good approximations to the PML distribution (or at least a locally optimal PML distribution), and, for the same level of accuracy, is two to three orders of magnitude faster than the former approach for practically relevant problem sizes. Moreover, the above free energy framework allows us to simplify some earlier proofs of properties of the PML distribution and to derive some new properties of the PML distribution, along with obtaining similar results for its Bethe approximation.</description><subject>Approximation algorithms</subject><subject>Approximation methods</subject><subject>Graphical models</subject><subject>Maximum likelihood estimation</subject><subject>Mercury (metals)</subject><subject>Minimization</subject><subject>Vectors</subject><issn>2157-8095</issn><issn>2157-8117</issn><isbn>9781467325806</isbn><isbn>1467325805</isbn><isbn>9781467325783</isbn><isbn>1467325791</isbn><isbn>9781467325790</isbn><isbn>1467325783</isbn><fulltext>true</fulltext><rsrctype>conference_proceeding</rsrctype><creationdate>2012</creationdate><recordtype>conference_proceeding</recordtype><sourceid>6IE</sourceid><recordid>eNpNkN1KxDAQheMfuKx9APEmL9Ca_6SXurhaWPDCer2kmwkbbbclzYK-vS2u4FycgfkOc4ZB6JaSglJS3ldvVV0wQlmhmOFKijOUldpQoTRnUht-jhaMSp0bSvXFf2aIuvxjpJTXKBvHDzKVnq1ygdb1HvAjpEntMMT-K3Q2hf6Ae4_n4WBTgnjAnZ3IscNt-IQ27PveYRfGFENznO036MrbdoTs1Jfoff1Ur17yzetztXrY5GFKS7kzjRWyBCm92XFHlVJCWeeUKInRxJOdtBSMJ5Y3hjfOK-G5kZSBU6wByZfo7ndvAIDtEKdr4_f29BX-A3acUfU</recordid><startdate>201207</startdate><enddate>201207</enddate><creator>Vontobel, P. O.</creator><general>IEEE</general><scope>6IE</scope><scope>6IH</scope><scope>CBEJK</scope><scope>RIE</scope><scope>RIO</scope></search><sort><creationdate>201207</creationdate><title>The Bethe approximation of the pattern maximum likelihood distribution</title><author>Vontobel, P. O.</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-i175t-d8ba459e55f8c3d166646add6490870f0c5a1e8f0a3b83bdf64f38512ed62be53</frbrgroupid><rsrctype>conference_proceedings</rsrctype><prefilter>conference_proceedings</prefilter><language>eng</language><creationdate>2012</creationdate><topic>Approximation algorithms</topic><topic>Approximation methods</topic><topic>Graphical models</topic><topic>Maximum likelihood estimation</topic><topic>Mercury (metals)</topic><topic>Minimization</topic><topic>Vectors</topic><toplevel>online_resources</toplevel><creatorcontrib>Vontobel, P. O.</creatorcontrib><collection>IEEE Electronic Library (IEL) Conference Proceedings</collection><collection>IEEE Proceedings Order Plan (POP) 1998-present by volume</collection><collection>IEEE Xplore All Conference Proceedings</collection><collection>IEEE Xplore (Online service)</collection><collection>IEEE Proceedings Order Plans (POP) 1998-present</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Vontobel, P. O.</au><format>book</format><genre>proceeding</genre><ristype>CONF</ristype><atitle>The Bethe approximation of the pattern maximum likelihood distribution</atitle><btitle>2012 IEEE International Symposium on Information Theory Proceedings</btitle><stitle>ISIT</stitle><date>2012-07</date><risdate>2012</risdate><spage>2012</spage><epage>2016</epage><pages>2012-2016</pages><issn>2157-8095</issn><eissn>2157-8117</eissn><isbn>9781467325806</isbn><isbn>1467325805</isbn><eisbn>9781467325783</eisbn><eisbn>1467325791</eisbn><eisbn>9781467325790</eisbn><eisbn>1467325783</eisbn><abstract>Among all memoryless source distributions, the pattern maximum likelihood (PML) distribution is the distribution which maximizes the probability that a memoryless source produces a string with a given pattern. Equivalently, the PML distribution maximizes the permanent of a certain non-negative matrix. We reformulate this maximization problem as a double minimization problem of a suitable Gibbs free energy function. Because finding the minimum of this function appears intractable for practically relevant problem sizes, one must look for tractable approximations. One approach is to approximately find a minimum (or at least a local minimum) of the Gibbs free energy function by applying an alternating minimization algorithm where the steps are based on quantities that are obtained by Markov chain Monte Carlo sampling. One can show that this approach is equivalent to an algorithm that was proposed by Orlitsky et al. An alternative approach is to replace the Gibbs free energy function by a tractable approximation like the Bethe free energy function and to apply an alternating minimization algorithm to this function. As it turns out, empirically, this latter approach gives very good approximations to the PML distribution (or at least a locally optimal PML distribution), and, for the same level of accuracy, is two to three orders of magnitude faster than the former approach for practically relevant problem sizes. Moreover, the above free energy framework allows us to simplify some earlier proofs of properties of the PML distribution and to derive some new properties of the PML distribution, along with obtaining similar results for its Bethe approximation.</abstract><pub>IEEE</pub><doi>10.1109/ISIT.2012.6283654</doi><tpages>5</tpages></addata></record> |
fulltext | fulltext_linktorsrc |
identifier | ISSN: 2157-8095 |
ispartof | 2012 IEEE International Symposium on Information Theory Proceedings, 2012, p.2012-2016 |
issn | 2157-8095 2157-8117 |
language | eng |
recordid | cdi_ieee_primary_6283654 |
source | IEEE Electronic Library (IEL) Conference Proceedings |
subjects | Approximation algorithms Approximation methods Graphical models Maximum likelihood estimation Mercury (metals) Minimization Vectors |
title | The Bethe approximation of the pattern maximum likelihood distribution |
url | http://sfxeu10.hosted.exlibrisgroup.com/loughborough?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-26T11%3A39%3A07IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-ieee_6IE&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=proceeding&rft.atitle=The%20Bethe%20approximation%20of%20the%20pattern%20maximum%20likelihood%20distribution&rft.btitle=2012%20IEEE%20International%20Symposium%20on%20Information%20Theory%20Proceedings&rft.au=Vontobel,%20P.%20O.&rft.date=2012-07&rft.spage=2012&rft.epage=2016&rft.pages=2012-2016&rft.issn=2157-8095&rft.eissn=2157-8117&rft.isbn=9781467325806&rft.isbn_list=1467325805&rft_id=info:doi/10.1109/ISIT.2012.6283654&rft.eisbn=9781467325783&rft.eisbn_list=1467325791&rft.eisbn_list=9781467325790&rft.eisbn_list=1467325783&rft_dat=%3Cieee_6IE%3E6283654%3C/ieee_6IE%3E%3Cgrp_id%3Ecdi_FETCH-LOGICAL-i175t-d8ba459e55f8c3d166646add6490870f0c5a1e8f0a3b83bdf64f38512ed62be53%3C/grp_id%3E%3Coa%3E%3C/oa%3E%3Curl%3E%3C/url%3E&rft_id=info:oai/&rft_id=info:pmid/&rft_ieee_id=6283654&rfr_iscdi=true |