Towards flatter loss surface via nonmonotonic learning rate scheduling

Whereas optimizing deep neural networks using stochastic gradient descent has shown great performances in practice, the rule for setting step size (i.e. learning rate) of gradient descent is not well studied. Although it appears that some intriguing learning rate rules such as ADAM (Kingma and Ba, 2014) have since been developed, they concentrated on improving convergence, not on improving generalization capabilities. Recently, the improved generalization property of the flat minima was revisited, and this research guides us towards promising solutions to many current optimization problems. In this paper, we analyze the flatness of loss surfaces through the lens of robustness to input perturbations and advocate that gradient descent should be guided to reach flatter region of loss surfaces to achieve generalization. Finally, we suggest a learning rate rule for escaping sharp regions of loss surfaces, and we demonstrate the capacity of our approach by performing numerous experiments.