Rényi Divergence Deep Mutual Learning

作     者：Huang, Weipeng Huang, Weipeng Fuzzy Tao, Junjie Deng, Changbo Fan, Ming Wan, Wenqiang Xiong, Qi Piao, Guangyuan 

作者机构：Tencent Security Big Data Lab Shenzhen China School of Software Engineering Xi’an Jiaotong University Xi’an China The MoEKLINNS Laboratory School of Cyber Science and Engineering Xi’an Jiaotong University Xi’an710049 China Department of Computer Science Maynooth University Maynooth Ireland 

出 版 物：《arXiv》 (arXiv)

年 卷 期：2022年

核心收录：

主　　题：Contrastive Learning 

摘      要：This paper revisits Deep Mutual Learning (DML), a simple yet effective computing paradigm. We propose using Rényi divergence instead of the Kullback–Leibler 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 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 the Rényi divergence, leading to further improvement in model generalization. © 2022, CC BY.