Timezone: »

 
DIMENSION-REDUCED ADAPTIVE GRADIENT METHOD
Jingyang Li · Pan Zhou · Kuangyu Ding · Kim-Chuan Toh · Yinyu Ye
Event URL: https://openreview.net/forum?id=6ofsu7wzSMm »
Adaptive gradient methods, such as Adam, have shown faster convergence speed than SGD across various kinds of network models at the expense of inferior generalization performance. In this work, we proposed a Dimension-Reduced Adaptive Gradient Method (DRAG) to eliminate the generalization gap. DRAG makes an elegant combination of SGD and Adam by adopting a trust-region like framework. We observe that 1) Adam adjusts stepsizes for each gradient coordinate according to some loss curvature, and indeed decomposes the $n$-dimensional gradient into $n$ standard basis directions to search; 2) SGD uniformly scales gradient for all gradient coordinates and actually has only one descent direction to minimize. Accordingly, DRAG reduces the high degree of freedom of Adam and also improves the flexibility of SGD via optimizing the loss along $k\ (\ll \! n)$ descent directions, e.g. the gradient direction and momentum direction used in this work. Then per iteration, DRAG finds the best stepsizes for $k$ descent directions by solving a trust-region subproblem whose computational overhead is negligible since the trust-region subproblem is low-dimensional, e.g. $k=2$ in this work. DRAG is compatible with the common deep learning training pipeline without introducing extra hyper-parameters and with negligible extra computation. Experimental results on representative benchmarks testify the fast convergence speed and also superior generalization of DRAG.

Author Information

Jingyang Li (National University of Singapore)
Pan Zhou (SEA AI Lab)

Currently, I am a senior Research Scientist in Sea AI Lab of Sea group. Before, I worked in Salesforce as a research scientist during 2019 to 2021. I completed my Ph.D. degree in 2019 at the National University of Singapore (NUS), fortunately advised by Prof. Jiashi Feng and Prof. Shuicheng Yan. Before studying in NUS, I graduated from Peking University (PKU) in 2016 and during this period, I was fortunately directed by Prof. Zhouchen Lin and Prof. Chao Zhang in ZERO Lab. During the research period, I also work closely with Prof. Xiaotong Yuan. I also spend several wonderful months in 2018 at Georgia Tech as visiting student hosted by Prof. Huan Xu.

Kuangyu Ding (National University of Singapore)
Kim-Chuan Toh
Yinyu Ye (Standord)

More from the Same Authors