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 2103.01516cs.LGcs.CRmath.OCstat.ML
收藏

私有随机凸优化:$\ell_1$几何中的最优速率

Private Stochastic Convex Optimization: Optimal Rates in $\ell_1$ Geometry

Stanford University(斯坦福大学);Apple(苹果公司);Tel Aviv University(特拉维夫大学);Google(谷歌公司)

Hilal Asi, Vitaly Feldman, Tomer Koren, Kunal Talwar

摘要:

在$\ell_1$ -有界域上的随机凸优化在机器学习应用(如LASSO)中是普遍存在的,但在使用差分隐私进行学习时仍然知之甚少。我们发现,对数因子的最佳过剩人口损失的任何$(\varepsilon,δ)$ -差分私人优化是$\sqrt{\log(d)/n} + \sqrt{d}/\varepsilon n.$的上限是基于一个新的算法,结合了迭代本地化方法的~ \citet{FeldmanKoTa20}与私人正则化镜像下降的新分析。它适用于$p\in [1,2]$的$\ell_p$有界域,并且最多查询$n^{3/2}$梯度,这比需要$n^2$梯度的$\ell_2$情况下的最佳先前已知算法有所改进。此外,我们表明,当损失函数满足额外的平滑性假设,过量损失的上限(对数因子)由$\sqrt{\log(d)/n} + (\log(d)/\varepsilon n)^{2/3}.$这界是通过一个新的方差减少版本的弗兰克-沃尔夫算法,只需要一个单一的数据通过。我们还表明,在这种情况下的下限是上述两个速率的最小值。

英文摘要:

Stochastic convex optimization over an $\ell_1$-bounded domain is ubiquitous in machine learning applications such as LASSO but remains poorly understood when learning with differential privacy. We show that, up to logarithmic factors the optimal excess population loss of any $(\varepsilon,δ)$-differentially private optimizer is $\sqrt{\log(d)/n} + \sqrt{d}/\varepsilon n.$ The upper bound is based on a new algorithm that combines the iterative localization approach of~\citet{FeldmanKoTa20} with a new analysis of private regularized mirror descent. It applies to $\ell_p$ bounded domains for $p\in [1,2]$ and queries at most $n^{3/2}$ gradients improving over the best previously known algorithm for the $\ell_2$ case which needs $n^2$ gradients. Further, we show that when the loss functions satisfy additional smoothness assumptions, the excess loss is upper bounded (up to logarithmic factors) by $\sqrt{\log(d)/n} + (\log(d)/\varepsilon n)^{2/3}.$ This bound is achieved by a new variance-reduced version of the Frank-Wolfe algorithm that requires just a single pass over the data. We also show that the lower bound in this case is the minimum of the two rates mentioned above.

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。

↑