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, which further improves generalized models.

翻译：本文重新审视了深度互学习（DML）这一简单而有效的计算范式。我们提出使用更灵活且可调的Rényi散度替代KL散度，以改进基础DML。该改进能在有限额外复杂度的情况下持续提升基础DML的性能。我们从理论上分析了所提范式的收敛特性，证明在非凸优化任务的最坏情形下，采用恒定学习率的随机梯度下降法会以$\mathcal{O}(1)$偏差收敛。这意味着学习过程会到达邻近局部最优解，但会在有界范围内持续搜索，这有助于缓解过拟合。最后，广泛的实验结果表明，结合DML与Rényi散度能进一步提升泛化模型性能。