arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2205.04583math.OC
收藏

带有随机Polyak步长的SGD动力学:真正自适应的变体及收敛到精确解

Dynamics of SGD with Stochastic Polyak Stepsizes: Truly Adaptive Variants and Convergence to Exact Solution

Antonio Orvieto, Simon Lacoste-Julien, Nicolas Loizou

AI总结:

针对随机Polyak步长(SPS)在非过参数化场景下需预知最优损失且仅收敛至邻域的缺点,提出新变体DecSPS,在无需预知参数及限制性假设下实现强凸优化精确收敛。

摘要:

最近,Loizou等人(2021)提出并分析了带有随机Polyak步长(SPS)的随机梯度下降(SGD)。所提出的SPS具有强大的收敛保证和有竞争力的性能;然而,在非过参数化场景下使用它有两个主要缺点:(i)它需要预先知道最优小批量损失,当不满足插值条件(例如,正则化目标)时这是不可用的,以及(ii)它仅保证收敛到解的邻域。在这项工作中,我们研究了配备随机Polyak步长新变体的SGD的动力学和收敛性质,并为原始SPS的两个缺点提供了解决方案。我们首先表明,对原始SPS的一个简单修改,即使用下界代替最优函数值,可以直接解决问题(i)。另一方面,解决问题(ii)更具挑战性,并让我们对该方法的行为获得了有价值的见解。我们表明,如果不满足插值,SPS与随机梯度之间的相关性会引入偏差,这有效地扭曲了极小值点附近的梯度信号期望,导致不收敛——即使在训练期间缩小步长。为了解决这个问题,我们提出了DecSPS,这是对SPS的一种新颖修改,它保证收敛到精确极小值点——无需预先知道问题参数。对于强凸优化问题,DecSPS是第一个收敛到精确解的随机自适应优化方法,无需像有界迭代/梯度这样的限制性假设。

英文摘要:

Recently, Loizou et al. (2021), proposed and analyzed stochastic gradient descent (SGD) with stochastic Polyak stepsize (SPS). The proposed SPS comes with strong convergence guarantees and competitive performance; however, it has two main drawbacks when it is used in non-over-parameterized regimes: (i) It requires a priori knowledge of the optimal mini-batch losses, which are not available when the interpolation condition is not satisfied (e.g., regularized objectives), and (ii) it guarantees convergence only to a neighborhood of the solution. In this work, we study the dynamics and the convergence properties of SGD equipped with new variants of the stochastic Polyak stepsize and provide solutions to both drawbacks of the original SPS. We first show that a simple modification of the original SPS that uses lower bounds instead of the optimal function values can directly solve issue (i). On the other hand, solving issue (ii) turns out to be more challenging and leads us to valuable insights into the method's behavior. We show that if interpolation is not satisfied, the correlation between SPS and stochastic gradients introduces a bias, which effectively distorts the expectation of the gradient signal near minimizers, leading to non-convergence - even if the stepsize is scaled down during training. To fix this issue, we propose DecSPS, a novel modification of SPS, which guarantees convergence to the exact minimizer - without a priori knowledge of the problem parameters. For strongly-convex optimization problems, DecSPS is the first stochastic adaptive optimization method that converges to the exact solution without restrictive assumptions like bounded iterates/gradients.

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。

↑