Learn2Hop: Learned Optimization on Rough Landscapes

Optimization of non-convex loss surfaces containing many local minima remains a critical problem in a variety of domains, including operations research, informatics, and material design. Yet, current techniques either require extremely high iteration counts or a large number of random restarts for g...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Merchant, Amil, Metz, Luke, Schoenholz, Sam, Cubuk, Ekin Dogus
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 Merchant, Amil
Metz, Luke
Schoenholz, Sam
Cubuk, Ekin Dogus
description Optimization of non-convex loss surfaces containing many local minima remains a critical problem in a variety of domains, including operations research, informatics, and material design. Yet, current techniques either require extremely high iteration counts or a large number of random restarts for good performance. In this work, we propose adapting recent developments in meta-learning to these many-minima problems by learning the optimization algorithm for various loss landscapes. We focus on problems from atomic structural optimization--finding low energy configurations of many-atom systems--including widely studied models such as bimetallic clusters and disordered silicon. We find that our optimizer learns a 'hopping' behavior which enables efficient exploration and improves the rate of low energy minima discovery. Finally, our learned optimizers show promising generalization with efficiency gains on never before seen tasks (e.g. new elements or compositions). Code will be made available shortly.
doi_str_mv 10.48550/arxiv.2107.09661
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2107_09661</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2107_09661</sourcerecordid><originalsourceid>FETCH-LOGICAL-a671-2c212bb5eab8aff038fcbf177b45f519db345d0f2a79402e164760b213dbbe23</originalsourceid><addsrcrecordid>eNotzs1Kw0AUBeDZdCGtD-DKvEDSuXf-EnelqBUCBXUf7s3MtAM2CUkV9enVWDhwzurwCXEDstClMXJN42f6KBCkK2RlLVwJVQcaO9z1w102z-Cz_XBOp_RN59R32W-e-_fDMaup81NLQ5hWYhHpbQrXl16Kl4f71-0ur_ePT9tNnZN1kGOLgMwmEJcUo1RlbDmCc6xNNFB5Vtp4GZFcpSUGsNpZyQjKMwdUS3H7_zqjm2FMJxq_mj98M-PVD6o2Po8</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Learn2Hop: Learned Optimization on Rough Landscapes</title><source>arXiv.org</source><creator>Merchant, Amil ; Metz, Luke ; Schoenholz, Sam ; Cubuk, Ekin Dogus</creator><creatorcontrib>Merchant, Amil ; Metz, Luke ; Schoenholz, Sam ; Cubuk, Ekin Dogus</creatorcontrib><description>Optimization of non-convex loss surfaces containing many local minima remains a critical problem in a variety of domains, including operations research, informatics, and material design. Yet, current techniques either require extremely high iteration counts or a large number of random restarts for good performance. In this work, we propose adapting recent developments in meta-learning to these many-minima problems by learning the optimization algorithm for various loss landscapes. We focus on problems from atomic structural optimization--finding low energy configurations of many-atom systems--including widely studied models such as bimetallic clusters and disordered silicon. We find that our optimizer learns a 'hopping' behavior which enables efficient exploration and improves the rate of low energy minima discovery. Finally, our learned optimizers show promising generalization with efficiency gains on never before seen tasks (e.g. new elements or compositions). Code will be made available shortly.</description><identifier>DOI: 10.48550/arxiv.2107.09661</identifier><language>eng</language><subject>Computer Science - Learning</subject><creationdate>2021-07</creationdate><rights>http://creativecommons.org/licenses/by/4.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,776,881</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2107.09661$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2107.09661$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Merchant, Amil</creatorcontrib><creatorcontrib>Metz, Luke</creatorcontrib><creatorcontrib>Schoenholz, Sam</creatorcontrib><creatorcontrib>Cubuk, Ekin Dogus</creatorcontrib><title>Learn2Hop: Learned Optimization on Rough Landscapes</title><description>Optimization of non-convex loss surfaces containing many local minima remains a critical problem in a variety of domains, including operations research, informatics, and material design. Yet, current techniques either require extremely high iteration counts or a large number of random restarts for good performance. In this work, we propose adapting recent developments in meta-learning to these many-minima problems by learning the optimization algorithm for various loss landscapes. We focus on problems from atomic structural optimization--finding low energy configurations of many-atom systems--including widely studied models such as bimetallic clusters and disordered silicon. We find that our optimizer learns a 'hopping' behavior which enables efficient exploration and improves the rate of low energy minima discovery. Finally, our learned optimizers show promising generalization with efficiency gains on never before seen tasks (e.g. new elements or compositions). Code will be made available shortly.</description><subject>Computer Science - Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2021</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotzs1Kw0AUBeDZdCGtD-DKvEDSuXf-EnelqBUCBXUf7s3MtAM2CUkV9enVWDhwzurwCXEDstClMXJN42f6KBCkK2RlLVwJVQcaO9z1w102z-Cz_XBOp_RN59R32W-e-_fDMaup81NLQ5hWYhHpbQrXl16Kl4f71-0ur_ePT9tNnZN1kGOLgMwmEJcUo1RlbDmCc6xNNFB5Vtp4GZFcpSUGsNpZyQjKMwdUS3H7_zqjm2FMJxq_mj98M-PVD6o2Po8</recordid><startdate>20210720</startdate><enddate>20210720</enddate><creator>Merchant, Amil</creator><creator>Metz, Luke</creator><creator>Schoenholz, Sam</creator><creator>Cubuk, Ekin Dogus</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20210720</creationdate><title>Learn2Hop: Learned Optimization on Rough Landscapes</title><author>Merchant, Amil ; Metz, Luke ; Schoenholz, Sam ; Cubuk, Ekin Dogus</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a671-2c212bb5eab8aff038fcbf177b45f519db345d0f2a79402e164760b213dbbe23</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2021</creationdate><topic>Computer Science - Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Merchant, Amil</creatorcontrib><creatorcontrib>Metz, Luke</creatorcontrib><creatorcontrib>Schoenholz, Sam</creatorcontrib><creatorcontrib>Cubuk, Ekin Dogus</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Merchant, Amil</au><au>Metz, Luke</au><au>Schoenholz, Sam</au><au>Cubuk, Ekin Dogus</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Learn2Hop: Learned Optimization on Rough Landscapes</atitle><date>2021-07-20</date><risdate>2021</risdate><abstract>Optimization of non-convex loss surfaces containing many local minima remains a critical problem in a variety of domains, including operations research, informatics, and material design. Yet, current techniques either require extremely high iteration counts or a large number of random restarts for good performance. In this work, we propose adapting recent developments in meta-learning to these many-minima problems by learning the optimization algorithm for various loss landscapes. We focus on problems from atomic structural optimization--finding low energy configurations of many-atom systems--including widely studied models such as bimetallic clusters and disordered silicon. We find that our optimizer learns a 'hopping' behavior which enables efficient exploration and improves the rate of low energy minima discovery. Finally, our learned optimizers show promising generalization with efficiency gains on never before seen tasks (e.g. new elements or compositions). Code will be made available shortly.</abstract><doi>10.48550/arxiv.2107.09661</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2107.09661
ispartof
issn
language eng
recordid cdi_arxiv_primary_2107_09661
source arXiv.org
subjects Computer Science - Learning
title Learn2Hop: Learned Optimization on Rough Landscapes
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-02-07T21%3A09%3A45IST&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=Learn2Hop:%20Learned%20Optimization%20on%20Rough%20Landscapes&rft.au=Merchant,%20Amil&rft.date=2021-07-20&rft_id=info:doi/10.48550/arxiv.2107.09661&rft_dat=%3Carxiv_GOX%3E2107_09661%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