Robust Training in High Dimensions via Block Coordinate Geometric Median Descent
Geometric median (\textsc{Gm}) is a classical method in statistics for achieving a robust estimation of the uncorrupted data; under gross corruption, it achieves the optimal breakdown point of 0.5. However, its computational complexity makes it infeasible for robustifying stochastic gradient descent...
Gespeichert in:
Hauptverfasser: | , , , , , |
---|---|
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 | Acharya, Anish Hashemi, Abolfazl Jain, Prateek Sanghavi, Sujay Dhillon, Inderjit S Topcu, Ufuk |
description | Geometric median (\textsc{Gm}) is a classical method in statistics for
achieving a robust estimation of the uncorrupted data; under gross corruption,
it achieves the optimal breakdown point of 0.5. However, its computational
complexity makes it infeasible for robustifying stochastic gradient descent
(SGD) for high-dimensional optimization problems. In this paper, we show that
by applying \textsc{Gm} to only a judiciously chosen block of coordinates at a
time and using a memory mechanism, one can retain the breakdown point of 0.5
for smooth non-convex problems, with non-asymptotic convergence rates
comparable to the SGD with \textsc{Gm}. |
doi_str_mv | 10.48550/arxiv.2106.08882 |
format | Article |
fullrecord | <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2106_08882</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2106_08882</sourcerecordid><originalsourceid>FETCH-LOGICAL-a672-65535e2f53f5b1eef37a6b8e805e22f06b3c5c99b71e902f12a7e6ca35d1b3c63</originalsourceid><addsrcrecordid>eNotz81OwzAQBGBfOKDCA3DCL5DgH-w4R0ihRSoCodyjjbMuKxobOaGCt6e0nEaakUb6GLuSorx1xogbyN-0L5UUthTOOXXOXt9S_zXNvM1AkeKWU-Rr2r7zJY0YJ0px4nsCfr9L_oM3KeWBIszIV5hGnDN5_owDQeRLnDzG-YKdBdhNePmfC9Y-PrTNuti8rJ6au00BtlKFNUYbVMHoYHqJGHQFtnfoxKFVQdhee-Pruq8k1kIFqaBC60GbQR4mqxfs-nR7JHWfmUbIP90frTvS9C8ICkle</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Robust Training in High Dimensions via Block Coordinate Geometric Median Descent</title><source>arXiv.org</source><creator>Acharya, Anish ; Hashemi, Abolfazl ; Jain, Prateek ; Sanghavi, Sujay ; Dhillon, Inderjit S ; Topcu, Ufuk</creator><creatorcontrib>Acharya, Anish ; Hashemi, Abolfazl ; Jain, Prateek ; Sanghavi, Sujay ; Dhillon, Inderjit S ; Topcu, Ufuk</creatorcontrib><description>Geometric median (\textsc{Gm}) is a classical method in statistics for
achieving a robust estimation of the uncorrupted data; under gross corruption,
it achieves the optimal breakdown point of 0.5. However, its computational
complexity makes it infeasible for robustifying stochastic gradient descent
(SGD) for high-dimensional optimization problems. In this paper, we show that
by applying \textsc{Gm} to only a judiciously chosen block of coordinates at a
time and using a memory mechanism, one can retain the breakdown point of 0.5
for smooth non-convex problems, with non-asymptotic convergence rates
comparable to the SGD with \textsc{Gm}.</description><identifier>DOI: 10.48550/arxiv.2106.08882</identifier><language>eng</language><subject>Computer Science - Distributed, Parallel, and Cluster Computing ; Computer Science - Learning ; Mathematics - Optimization and Control ; Statistics - Machine Learning</subject><creationdate>2021-06</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>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,780,885</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2106.08882$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2106.08882$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Acharya, Anish</creatorcontrib><creatorcontrib>Hashemi, Abolfazl</creatorcontrib><creatorcontrib>Jain, Prateek</creatorcontrib><creatorcontrib>Sanghavi, Sujay</creatorcontrib><creatorcontrib>Dhillon, Inderjit S</creatorcontrib><creatorcontrib>Topcu, Ufuk</creatorcontrib><title>Robust Training in High Dimensions via Block Coordinate Geometric Median Descent</title><description>Geometric median (\textsc{Gm}) is a classical method in statistics for
achieving a robust estimation of the uncorrupted data; under gross corruption,
it achieves the optimal breakdown point of 0.5. However, its computational
complexity makes it infeasible for robustifying stochastic gradient descent
(SGD) for high-dimensional optimization problems. In this paper, we show that
by applying \textsc{Gm} to only a judiciously chosen block of coordinates at a
time and using a memory mechanism, one can retain the breakdown point of 0.5
for smooth non-convex problems, with non-asymptotic convergence rates
comparable to the SGD with \textsc{Gm}.</description><subject>Computer Science - Distributed, Parallel, and Cluster Computing</subject><subject>Computer Science - Learning</subject><subject>Mathematics - Optimization and Control</subject><subject>Statistics - Machine Learning</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2021</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotz81OwzAQBGBfOKDCA3DCL5DgH-w4R0ihRSoCodyjjbMuKxobOaGCt6e0nEaakUb6GLuSorx1xogbyN-0L5UUthTOOXXOXt9S_zXNvM1AkeKWU-Rr2r7zJY0YJ0px4nsCfr9L_oM3KeWBIszIV5hGnDN5_owDQeRLnDzG-YKdBdhNePmfC9Y-PrTNuti8rJ6au00BtlKFNUYbVMHoYHqJGHQFtnfoxKFVQdhee-Pruq8k1kIFqaBC60GbQR4mqxfs-nR7JHWfmUbIP90frTvS9C8ICkle</recordid><startdate>20210616</startdate><enddate>20210616</enddate><creator>Acharya, Anish</creator><creator>Hashemi, Abolfazl</creator><creator>Jain, Prateek</creator><creator>Sanghavi, Sujay</creator><creator>Dhillon, Inderjit S</creator><creator>Topcu, Ufuk</creator><scope>AKY</scope><scope>AKZ</scope><scope>EPD</scope><scope>GOX</scope></search><sort><creationdate>20210616</creationdate><title>Robust Training in High Dimensions via Block Coordinate Geometric Median Descent</title><author>Acharya, Anish ; Hashemi, Abolfazl ; Jain, Prateek ; Sanghavi, Sujay ; Dhillon, Inderjit S ; Topcu, Ufuk</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a672-65535e2f53f5b1eef37a6b8e805e22f06b3c5c99b71e902f12a7e6ca35d1b3c63</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2021</creationdate><topic>Computer Science - Distributed, Parallel, and Cluster Computing</topic><topic>Computer Science - Learning</topic><topic>Mathematics - Optimization and Control</topic><topic>Statistics - Machine Learning</topic><toplevel>online_resources</toplevel><creatorcontrib>Acharya, Anish</creatorcontrib><creatorcontrib>Hashemi, Abolfazl</creatorcontrib><creatorcontrib>Jain, Prateek</creatorcontrib><creatorcontrib>Sanghavi, Sujay</creatorcontrib><creatorcontrib>Dhillon, Inderjit S</creatorcontrib><creatorcontrib>Topcu, Ufuk</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv Mathematics</collection><collection>arXiv Statistics</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Acharya, Anish</au><au>Hashemi, Abolfazl</au><au>Jain, Prateek</au><au>Sanghavi, Sujay</au><au>Dhillon, Inderjit S</au><au>Topcu, Ufuk</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Robust Training in High Dimensions via Block Coordinate Geometric Median Descent</atitle><date>2021-06-16</date><risdate>2021</risdate><abstract>Geometric median (\textsc{Gm}) is a classical method in statistics for
achieving a robust estimation of the uncorrupted data; under gross corruption,
it achieves the optimal breakdown point of 0.5. However, its computational
complexity makes it infeasible for robustifying stochastic gradient descent
(SGD) for high-dimensional optimization problems. In this paper, we show that
by applying \textsc{Gm} to only a judiciously chosen block of coordinates at a
time and using a memory mechanism, one can retain the breakdown point of 0.5
for smooth non-convex problems, with non-asymptotic convergence rates
comparable to the SGD with \textsc{Gm}.</abstract><doi>10.48550/arxiv.2106.08882</doi><oa>free_for_read</oa></addata></record> |
fulltext | fulltext_linktorsrc |
identifier | DOI: 10.48550/arxiv.2106.08882 |
ispartof | |
issn | |
language | eng |
recordid | cdi_arxiv_primary_2106_08882 |
source | arXiv.org |
subjects | Computer Science - Distributed, Parallel, and Cluster Computing Computer Science - Learning Mathematics - Optimization and Control Statistics - Machine Learning |
title | Robust Training in High Dimensions via Block Coordinate Geometric Median Descent |
url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2024-12-25T05%3A12%3A54IST&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=Robust%20Training%20in%20High%20Dimensions%20via%20Block%20Coordinate%20Geometric%20Median%20Descent&rft.au=Acharya,%20Anish&rft.date=2021-06-16&rft_id=info:doi/10.48550/arxiv.2106.08882&rft_dat=%3Carxiv_GOX%3E2106_08882%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 |