R\'{e}nyi Divergence Deep Mutual Learning
This paper revisits Deep Mutual Learning (DML), a simple yet effective computing paradigm. We propose using R\'{e}nyi divergence instead of the KL divergence, which is more flexible and tunable, to improve vanilla DML. This modification is able to consistently improve performance over vanilla D...
Saved in:
Main Authors | , , , , , , |
---|---|
Format | Journal Article |
Language | English |
Published |
13.09.2022
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Abstract | This paper revisits Deep Mutual Learning (DML), a simple yet effective
computing paradigm. We propose using R\'{e}nyi divergence instead of the KL
divergence, which is more flexible and tunable, to improve vanilla DML. This
modification is able to consistently improve performance over vanilla DML with
limited additional complexity. The convergence properties of the proposed
paradigm are analyzed theoretically, and Stochastic Gradient Descent with a
constant learning rate is shown to converge with $\mathcal{O}(1)$-bias in the
worst case scenario for nonconvex optimization tasks. That is, learning will
reach nearby local optima but continue searching within a bounded scope, which
may help mitigate overfitting. Finally, our extensive empirical results
demonstrate the advantage of combining DML and R\'{e}nyi divergence, leading to
further improvement in model generalization. |
---|---|
AbstractList | This paper revisits Deep Mutual Learning (DML), a simple yet effective
computing paradigm. We propose using R\'{e}nyi divergence instead of the KL
divergence, which is more flexible and tunable, to improve vanilla DML. This
modification is able to consistently improve performance over vanilla DML with
limited additional complexity. The convergence properties of the proposed
paradigm are analyzed theoretically, and Stochastic Gradient Descent with a
constant learning rate is shown to converge with $\mathcal{O}(1)$-bias in the
worst case scenario for nonconvex optimization tasks. That is, learning will
reach nearby local optima but continue searching within a bounded scope, which
may help mitigate overfitting. Finally, our extensive empirical results
demonstrate the advantage of combining DML and R\'{e}nyi divergence, leading to
further improvement in model generalization. |
Author | Piao, Guangyuan Deng, Changbo Huang, Weipeng Tao, Junjie Wan, Wenqiang Xiong, Qi Fan, Ming |
Author_xml | – sequence: 1 givenname: Weipeng surname: Huang fullname: Huang, Weipeng – sequence: 2 givenname: Junjie surname: Tao fullname: Tao, Junjie – sequence: 3 givenname: Changbo surname: Deng fullname: Deng, Changbo – sequence: 4 givenname: Ming surname: Fan fullname: Fan, Ming – sequence: 5 givenname: Wenqiang surname: Wan fullname: Wan, Wenqiang – sequence: 6 givenname: Qi surname: Xiong fullname: Xiong, Qi – sequence: 7 givenname: Guangyuan surname: Piao fullname: Piao, Guangyuan |
BackLink | https://doi.org/10.48550/arXiv.2209.05732$$DView paper in arXiv |
BookMark | eNrjYmDJy89LZWCQNDTQM7EwNTXQTyyqyCzTMzIysNQzMDU3NuJk0AyKUa9Orc2rzFRwySxLLUpPzUtOVXBJTS1Q8C0tKU3MUfBJTSzKy8xL52FgTUvMKU7lhdLcDPJuriHOHrpgQ-MLijJzE4sq40GGx4MNNyasAgC9ni-Z |
ContentType | Journal Article |
Copyright | http://creativecommons.org/licenses/by/4.0 |
Copyright_xml | – notice: http://creativecommons.org/licenses/by/4.0 |
DBID | AKY GOX |
DOI | 10.48550/arxiv.2209.05732 |
DatabaseName | arXiv Computer Science arXiv.org |
DatabaseTitleList | |
Database_xml | – sequence: 1 dbid: GOX name: arXiv.org url: http://arxiv.org/find sourceTypes: Open Access Repository |
DeliveryMethod | fulltext_linktorsrc |
ExternalDocumentID | 2209_05732 |
GroupedDBID | AKY GOX |
ID | FETCH-arxiv_primary_2209_057323 |
IEDL.DBID | GOX |
IngestDate | Fri Sep 20 18:52:46 EDT 2024 |
IsDoiOpenAccess | true |
IsOpenAccess | true |
IsPeerReviewed | false |
IsScholarly | false |
Language | English |
LinkModel | DirectLink |
MergedId | FETCHMERGED-arxiv_primary_2209_057323 |
OpenAccessLink | https://arxiv.org/abs/2209.05732 |
ParticipantIDs | arxiv_primary_2209_05732 |
PublicationCentury | 2000 |
PublicationDate | 2022-09-13 |
PublicationDateYYYYMMDD | 2022-09-13 |
PublicationDate_xml | – month: 09 year: 2022 text: 2022-09-13 day: 13 |
PublicationDecade | 2020 |
PublicationYear | 2022 |
Score | 3.702938 |
SecondaryResourceType | preprint |
Snippet | This paper revisits Deep Mutual Learning (DML), a simple yet effective
computing paradigm. We propose using R\'{e}nyi divergence instead of the KL
divergence,... |
SourceID | arxiv |
SourceType | Open Access Repository |
SubjectTerms | Computer Science - Artificial Intelligence Computer Science - Learning |
Title | R\'{e}nyi Divergence Deep Mutual Learning |
URI | https://arxiv.org/abs/2209.05732 |
hasFullText | 1 |
inHoldings | 1 |
isFullTextHit | |
isPrint | |
link | http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwY2BQSTI3STE2M0zUtbRINNM1STFK1k1KMk8EZjzzlKRkQ0Nj0xTwAlk_M49QE68I0wgmBgXYXpjEoorMMsj5wEnF-kZGoOMkTc2NgYUss5ERaMmWu38EZHISfBQXVD1CHbCNCRZCqiTcBBn4oa07BUdIdAgxMKXmiTBoBsWoV6fW5lVmKriAVkGAj79UcElNLVDwLQVt31CAHnKaLsog7-Ya4uyhCzY8vgByEkQ8yN54sL3GYgwswP56qgSDQoqFmVlikkGiqVFymolZcmJiamJiUlqSRap5mmEasHskySCByxQp3FLSDFxGoJX3oNsLjGUYWEqKSlNlgfVhSZIcOFAAIcBkuw |
link.rule.ids | 228,230,786,891 |
linkProvider | Cornell University |
openUrl | ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=R%5C%27%7Be%7Dnyi+Divergence+Deep+Mutual+Learning&rft.au=Huang%2C+Weipeng&rft.au=Tao%2C+Junjie&rft.au=Deng%2C+Changbo&rft.au=Fan%2C+Ming&rft.date=2022-09-13&rft_id=info:doi/10.48550%2Farxiv.2209.05732&rft.externalDocID=2209_05732 |