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 2010.02479math.OCcs.AIcs.LG
收藏

一种基于比较的一位梯度估计器

A One-bit, Comparison-Based Gradient Estimator

University of California, Los Angeles(加州大学洛杉矶分校);Damo Academy, Alibaba US(阿里巴巴美国达摩院);Xmotors AI(Xmotors人工智能公司)

HanQin Cai, Daniel Mckenzie, Wotao Yin, Zhenliang Zhang

摘要:

我们研究零阶优化的凸函数,我们进一步假设的功能评估是不可用的。相反,人们只能访问$\textit{comparison oracle}$,给定两个点$x$和$y$,它返回单个比特的信息,指示哪个点具有更大的函数值,$f(x)$或$f(y)$。通过将梯度视为待恢复的未知信号,我们展示了如何使用来自一位压缩感知的工具来构建归一化梯度的鲁棒性和可靠性估计。然后,我们提出了一个算法,创造SCOBO,使用这个估计在梯度下降方案。我们表明,当$f(x)$有一些低维结构,可以利用,SCOBO优于国家的最先进的查询复杂性。我们的理论主张得到了广泛的数值实验的验证。

英文摘要:

We study zeroth-order optimization for convex functions where we further assume that function evaluations are unavailable. Instead, one only has access to a $\textit{comparison oracle}$, which given two points $x$ and $y$ returns a single bit of information indicating which point has larger function value, $f(x)$ or $f(y)$. By treating the gradient as an unknown signal to be recovered, we show how one can use tools from one-bit compressed sensing to construct a robust and reliable estimator of the normalized gradient. We then propose an algorithm, coined SCOBO, that uses this estimator within a gradient descent scheme. We show that when $f(x)$ has some low dimensional structure that can be exploited, SCOBO outperforms the state-of-the-art in terms of query complexity. Our theoretical claims are verified by extensive numerical experiments.

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。

↑