Hutch++: Optimal Stochastic Trace Estimation.

Hutch++: Optimal Stochastic Trace Estimation.
复制标题

DOI:
10.1137/1.9781611976496.16
复制
发表时间:
2021-01
期刊:
Proceedings of the SIAM Symposium on Simplicity in Algorithms (SOSA)
影响因子:
--
通讯作者:
Woodruff DP
Woodruff DP
中科院分区:
其他
文献类型:
--
作者:
Meyer RA;Musco C;Musco C;Woodruff DP

文献摘要

参考文献

被引文献

相似文献

我们研究了矩阵A的迹估计问题,该问题只能通过矩阵-向量乘法来获得。介绍了一种新的随机化算法Hutch++,该算法仅用O(1/ε)个矩阵向量积即可计算出任意半正定(PsD)A的(1±ε)个tr(A)逼近.这改进了普遍存在的Hutchinson估计,它需要O(1/ε2)个矩阵向量乘积。我们的方法基于一种简单的技术,可以使用低阶近似步骤来降低Hutchinson估计的方差,并且易于实现和分析。此外,我们证明了当查询可以自适应地选择时,Hutch++算法的复杂度在所有矩阵-向量查询算法中是最优的,直到对数倍。实验表明,该方法的性能明显优于Hutchinson的方法。虽然我们的理论要求A是半正定的,但经验收益扩展到了涉及非PSD矩阵的应用,例如网络中的三角估计。
We study the problem of estimating the trace of a matrix A that can only be accessed through matrix-vector multiplication. We introduce a new randomized algorithm, Hutch++, which computes a (1 ± ε) approximation to tr(A) for any positive semidefinite (PSD) A using just O(1/ε) matrix-vector products. This improves on the ubiquitous Hutchinson’s estimator, which requires O(1/ε2) matrix-vector products. Our approach is based on a simple technique for reducing the variance of Hutchinson’s estimator using a low-rank approximation step, and is easy to implement and analyze. Moreover, we prove that, up to a logarithmic factor, the complexity of Hutch++ is optimal amongst all matrix-vector query algorithms, even when queries can be chosen adaptively. We show that it significantly outperforms Hutchinson’s method in experiments. While our theory requires A to be positive semidefinite, empirical gains extend to applications involving non-PSD matrices, such as triangle estimation in networks.
DOI: 10.1137/120861072
发表时间: 2012-01-01
影响因子: 1.6
作者:
Chakrabarti, Amit;Regev, Oded
通讯作者: Regev, Oded
DOI: 10.1016/s0022-0000(03)00025-4
发表时间: 2003-06-01
影响因子: 1.1
作者:
Achlioptas, D
通讯作者: Achlioptas, D
DOI: 10.1007/s00440-014-0562-z
发表时间: 2015-04-01
影响因子: 2
作者:
Cai T;Ma Z;Wu Y
通讯作者: Wu Y
DOI: 10.1080/03610919008812866
发表时间: 1990-01-01
影响因子: 0.9
作者:
HUTCHINSON, MF
通讯作者: HUTCHINSON, MF
DOI: 10.1016/j.laa.2007.06.020
发表时间: 2007-11-01
影响因子: 1.1
作者:
de la Pena, Jose Antonio;Gutman, Ivan;Rada, Juan
通讯作者: Rada, Juan