Last iterate convergence of SGD for Least-Squares in the Interpolation regime

Last iterate convergence of SGD for Least-Squares in the Interpolation regime
复制标题

插值体系中最小二乘法 SGD 的最后一次迭代收敛

DOI:
--
复制
发表时间:
2021
期刊:
Neural Information Processing Systems
影响因子:
--
通讯作者:
Nicolas Flammarion
Nicolas Flammarion
中科院分区:
--
文献类型:
--
作者:
Aditya Varre;Loucas Pillaud;Nicolas Flammarion

文献摘要

参考文献

被引文献

相似文献

由于神经网络具有良好的数据拟合能力和泛化能力,我们研究了基本最小二乘设置中的无噪声模型。我们假设一个最优的预测器完美地拟合输入和输出$langle heta_*, phi(X) angle = Y$,其中$phi(X)$代表一个可能无限维的非线性特征映射。为了解决这一问题,我们考虑了常步长随机梯度下降(SGD)的最后一次迭代给出的估计量。在这方面,我们的贡献有两方面:(i)从(随机)优化的角度来看,我们展示了一个原型问题,在这个问题中,我们可以明确地显示出具有恒定步长的非强凸问题的SGD最终迭代的收敛性,而通常的结果使用某种形式的平均值;(ii)从统计的角度来看,我们在过度参数化的情况下给出了显式的非渐近收敛率,并利用问题的细粒度参数化来展示可以比$O(1/T)$更快的多项式速率。建立了与再现核希尔伯特空间的联系。
Motivated by the recent successes of neural networks that have the ability to fit the data perfectly and generalize well, we study the noiseless model in the fundamental least-squares setup. We assume that an optimum predictor fits perfectly inputs and outputs $langle heta_* , phi(X) angle = Y$, where $phi(X)$ stands for a possibly infinite dimensional non-linear feature map. To solve this problem, we consider the estimator given by the last iterate of stochastic gradient descent (SGD) with constant step-size. In this context, our contribution is two fold: (i) from a (stochastic) optimization perspective, we exhibit an archetypal problem where we can show explicitly the convergence of SGD final iterate for a non-strongly convex problem with constant step-size whereas usual results use some form of average and (ii) from a statistical perspective, we give explicit non-asymptotic convergence rates in the over-parameterized setting and leverage a fine-grained parameterization of the problem to exhibit polynomial rates that can be faster than $O(1/T)$. The link with reproducing kernel Hilbert spaces is established.
DOI: --
发表时间: 2019-05
期刊: ArXiv
影响因子: --
作者:
Kwang-Sung Jun;Ashok Cutkosky;Francesco Orabona
通讯作者: Kwang-Sung Jun;Ashok Cutkosky;Francesco Orabona
DOI: 10.1073/pnas.1903070116
发表时间: 2019-08-06
影响因子: 11.1
作者:
Belkin, Mikhail;Hsu, Daniel;Mandal, Soumik
通讯作者: Mandal, Soumik