ADASHIFT: DECORRELATION AND CONVERGENCE OF ADAPTIVE LEARNING RATE METHODS
1INTRODUCTION
词:1.respect to 关于 2.exponential 指数 3.alleviate 减轻 4.oscillation 震荡 5.counterexamples 反例 6.violation 违反
7.subsequently 随后
短语:1.the key issue。。。。。 lies in the 关键问题在于。。。 2.maintaining a decent performance 保持良好的表现
5.EXPERIMENTS
compare them with Adam, AMSGrad and SGD.The anonymous code is provided at http://bit.ly/2NDXX6x..
词:slightly 略微