Rényi Divergence Deep Mutual Learning ...
This paper revisits Deep Mutual Learning (DML), a simple yet effective computing paradigm. We propose using Rényi 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 l...
Main Authors: | , , , , , , |
---|---|
Format: | Report |
Language: | unknown |
Published: |
arXiv
2022
|
Subjects: | |
Online Access: | https://dx.doi.org/10.48550/arxiv.2209.05732 https://arxiv.org/abs/2209.05732 |
id |
ftdatacite:10.48550/arxiv.2209.05732 |
---|---|
record_format |
openpolar |
spelling |
ftdatacite:10.48550/arxiv.2209.05732 2024-10-13T14:06:50+00:00 Rényi Divergence Deep Mutual Learning ... Huang, Weipeng Tao, Junjie Deng, Changbo Fan, Ming Wan, Wenqiang Xiong, Qi Piao, Guangyuan 2022 https://dx.doi.org/10.48550/arxiv.2209.05732 https://arxiv.org/abs/2209.05732 unknown arXiv Creative Commons Attribution 4.0 International https://creativecommons.org/licenses/by/4.0/legalcode cc-by-4.0 Machine Learning cs.LG Artificial Intelligence cs.AI FOS: Computer and information sciences CreativeWork article Preprint Article 2022 ftdatacite https://doi.org/10.48550/arxiv.2209.05732 2024-10-01T11:29:14Z This paper revisits Deep Mutual Learning (DML), a simple yet effective computing paradigm. We propose using Rényi 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 Rényi divergence, leading to further improvement in model generalization. ... Report DML DataCite |
institution |
Open Polar |
collection |
DataCite |
op_collection_id |
ftdatacite |
language |
unknown |
topic |
Machine Learning cs.LG Artificial Intelligence cs.AI FOS: Computer and information sciences |
spellingShingle |
Machine Learning cs.LG Artificial Intelligence cs.AI FOS: Computer and information sciences Huang, Weipeng Tao, Junjie Deng, Changbo Fan, Ming Wan, Wenqiang Xiong, Qi Piao, Guangyuan Rényi Divergence Deep Mutual Learning ... |
topic_facet |
Machine Learning cs.LG Artificial Intelligence cs.AI FOS: Computer and information sciences |
description |
This paper revisits Deep Mutual Learning (DML), a simple yet effective computing paradigm. We propose using Rényi 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 Rényi divergence, leading to further improvement in model generalization. ... |
format |
Report |
author |
Huang, Weipeng Tao, Junjie Deng, Changbo Fan, Ming Wan, Wenqiang Xiong, Qi Piao, Guangyuan |
author_facet |
Huang, Weipeng Tao, Junjie Deng, Changbo Fan, Ming Wan, Wenqiang Xiong, Qi Piao, Guangyuan |
author_sort |
Huang, Weipeng |
title |
Rényi Divergence Deep Mutual Learning ... |
title_short |
Rényi Divergence Deep Mutual Learning ... |
title_full |
Rényi Divergence Deep Mutual Learning ... |
title_fullStr |
Rényi Divergence Deep Mutual Learning ... |
title_full_unstemmed |
Rényi Divergence Deep Mutual Learning ... |
title_sort |
rényi divergence deep mutual learning ... |
publisher |
arXiv |
publishDate |
2022 |
url |
https://dx.doi.org/10.48550/arxiv.2209.05732 https://arxiv.org/abs/2209.05732 |
genre |
DML |
genre_facet |
DML |
op_rights |
Creative Commons Attribution 4.0 International https://creativecommons.org/licenses/by/4.0/legalcode cc-by-4.0 |
op_doi |
https://doi.org/10.48550/arxiv.2209.05732 |
_version_ |
1812813051852750848 |