arXivDaily arXiv每日学术速递 周一至周五更新
北美退伍军人节(Veterans Day),12 日无更新
arXiv 2011.12547cs.LGstat.ML
收藏

深度线性网络在大学习率阶段的内隐偏差

Implicit bias of deep linear networks in the large learning rate phase

University of Technology Sydney(悉尼科技大学);Northwestern University(西北大学)

Wei Huang, Weitao Du, Richard Yi Da Xu, Chunrui Liu

摘要:

大多数解释深度学习中正则化效应的理论研究都只关注具有足够小的学习率甚至梯度流(无穷小的学习率)的梯度下降。然而,这些研究忽略了在大多数实际应用中应用的相当大的学习率。在这项工作中,我们使用大学习率机制中的逻辑损失来表征深度线性网络对二进制分类的隐式偏差效应,这受到Lewkowycz等人的开创性工作的启发。[26]在平方损失的回归设置中。他们发现了一个具有大步长的学习率制度,称为弹射阶段,在训练的早期阶段,损失增加,最终收敛到最小值,比小学习率制度中发现的更平坦。我们声称,根据分离条件的数据,梯度下降迭代将收敛到一个平坦的最小值在弹射阶段。我们严格证明了这一说法的假设下,退化的数据,克服了困难的非常数Hessian的logistic损失,并进一步刻画了行为的损失和Hessian的不可分数据。最后,我们证明了在弹射阶段,随着学习率的非可分离数据所跨越的空间中的平坦的最小值可以导致更好的经验推广。

英文摘要:

Most theoretical studies explaining the regularization effect in deep learning have only focused on gradient descent with a sufficient small learning rate or even gradient flow (infinitesimal learning rate). Such researches, however, have neglected a reasonably large learning rate applied in most practical applications. In this work, we characterize the implicit bias effect of deep linear networks for binary classification using the logistic loss in the large learning rate regime, inspired by the seminal work by Lewkowycz et al. [26] in a regression setting with squared loss. They found a learning rate regime with a large stepsize named the catapult phase, where the loss grows at the early stage of training and eventually converges to a minimum that is flatter than those found in the small learning rate regime. We claim that depending on the separation conditions of data, the gradient descent iterates will converge to a flatter minimum in the catapult phase. We rigorously prove this claim under the assumption of degenerate data by overcoming the difficulty of the non-constant Hessian of logistic loss and further characterize the behavior of loss and Hessian for non-separable data. Finally, we demonstrate that flatter minima in the space spanned by non-separable data along with the learning rate in the catapult phase can lead to better generalization empirically.

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。

↑