arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2105.09433cs.LGstat.ML
收藏

带Lewis加权子抽样的L1回归

L1 Regression with Lewis Weights Subsampling

UT Austin(德克萨斯大学奥斯汀分校)

Aditya Parulekar, Advait Parulekar, Eric Price

摘要:

我们考虑的问题,找到一个近似的解决方案,$\ell_1$回归,而只观察到少量的标签。给定一个$n \times d$未标记的数据矩阵$X$,我们必须选择一个小的$m \ll n$行集合来观察其标记,然后输出一个估计$\widehatβ$,其对原始问题的误差在最优的$1 + \varepsilon$因子内。我们表明,从$X$根据其刘易斯权重和输出的经验极小成功概率$1-δ$为$m > O(\frac{1}{\varepsilon^2} d \log \frac{d}{\varepsilon δ})$。这类似于根据$\ell_2$回归的杠杆分数进行采样的性能,但对$δ$的依赖性指数更好。我们也给出了相应的$Ω(\frac{d}{\varepsilon^2} + (d + \frac{1}{\varepsilon^2}) \log\frac{1}δ)$的下界。

英文摘要:

We consider the problem of finding an approximate solution to $\ell_1$ regression while only observing a small number of labels. Given an $n \times d$ unlabeled data matrix $X$, we must choose a small set of $m \ll n$ rows to observe the labels of, then output an estimate $\widehatβ$ whose error on the original problem is within a $1 + \varepsilon$ factor of optimal. We show that sampling from $X$ according to its Lewis weights and outputting the empirical minimizer succeeds with probability $1-δ$ for $m > O(\frac{1}{\varepsilon^2} d \log \frac{d}{\varepsilon δ})$. This is analogous to the performance of sampling according to leverage scores for $\ell_2$ regression, but with exponentially better dependence on $δ$. We also give a corresponding lower bound of $Ω(\frac{d}{\varepsilon^2} + (d + \frac{1}{\varepsilon^2}) \log\frac{1}δ)$.

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。

↑