Closing the Generalization Gap of Adaptive Gradient Methods in Training Deep Neural Networks
This addresses a key problem for deep learning practitioners by potentially enabling faster training without sacrificing generalization, though it is incremental as it builds on existing methods.
The paper tackles the generalization gap of adaptive gradient methods like Adam and Amsgrad compared to SGD in training deep neural networks, proposing a new algorithm that unifies them with SGD to achieve fast convergence while matching SGD's generalization, as shown in experiments on standard benchmarks.
Adaptive gradient methods, which adopt historical gradient information to automatically adjust the learning rate, despite the nice property of fast convergence, have been observed to generalize worse than stochastic gradient descent (SGD) with momentum in training deep neural networks. This leaves how to close the generalization gap of adaptive gradient methods an open problem. In this work, we show that adaptive gradient methods such as Adam, Amsgrad, are sometimes "over adapted". We design a new algorithm, called Partially adaptive momentum estimation method, which unifies the Adam/Amsgrad with SGD by introducing a partial adaptive parameter $p$, to achieve the best from both worlds. We also prove the convergence rate of our proposed algorithm to a stationary point in the stochastic nonconvex optimization setting. Experiments on standard benchmarks show that our proposed algorithm can maintain a fast convergence rate as Adam/Amsgrad while generalizing as well as SGD in training deep neural networks. These results would suggest practitioners pick up adaptive gradient methods once again for faster training of deep neural networks.