Rényi Divergence Deep Mutual Learning
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2022
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910608368074752 |
|---|---|
| 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 |
| contents | 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 | Preprint |
| id |
arxiv_https___arxiv_org_abs_2209_05732 |
| institution | arXiv |
| publishDate | 2022 |
| record_format | arxiv |
| spellingShingle | Rényi Divergence Deep Mutual Learning Huang, Weipeng Tao, Junjie Deng, Changbo Fan, Ming Wan, Wenqiang Xiong, Qi Piao, Guangyuan Machine Learning Artificial Intelligence 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. |
| title | Rényi Divergence Deep Mutual Learning |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2209.05732 |