arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1908.03265cs.LGcs.CLstat.ML
收藏

关于自适应学习率的方差及超越

On the Variance of the Adaptive Learning Rate and Beyond

University of Illinois, Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校);Georgia Tech(佐治亚理工学院);Microsoft Dynamics 365 AI(微软 Dynamics 365 AI);Microsoft Research(微软研究院)

Liyuan Liu, Haoming Jiang, Pengcheng He, Weizhu Chen, Xiaodong Liu, Jianfeng Gao, Jiawei Han

摘要:

学习率预热启发式在稳定训练、加速收敛和提高自适应随机优化算法(如RMSprop和Adam)的泛化能力方面取得了显著的成功。在此,我们对其机理进行了详细的研究。继续研究预热背后的理论,我们发现了自适应学习率的问题(即,它在早期阶段有问题的大方差),建议预热工程作为一种方差减少技术,并提供经验和理论证据来验证我们的假设。我们进一步提出了RAdam,亚当的一个新的变种,通过引入一个术语来纠正自适应学习率的方差。在图像分类、语言建模和神经机器翻译上的大量实验结果验证了我们的直觉,并证明了我们所提出的方法的有效性和鲁棒性。所有实现都可以在https://github.com/LiyuanLucasLiu/RAdam上获得。

英文摘要:

The learning rate warmup heuristic achieves remarkable success in stabilizing training, accelerating convergence and improving generalization for adaptive stochastic optimization algorithms like RMSprop and Adam. Here, we study its mechanism in details. Pursuing the theory behind warmup, we identify a problem of the adaptive learning rate (i.e., it has problematically large variance in the early stage), suggest warmup works as a variance reduction technique, and provide both empirical and theoretical evidence to verify our hypothesis. We further propose RAdam, a new variant of Adam, by introducing a term to rectify the variance of the adaptive learning rate. Extensive experimental results on image classification, language modeling, and neural machine translation verify our intuition and demonstrate the effectiveness and robustness of our proposed method. All implementations are available at: https://github.com/LiyuanLucasLiu/RAdam.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑