An O(N) algorithm for computing expectation of N-dimensional truncated multi-variate normal distribution I: fundamentals
An O(N) algorithm for computing expectation of N-dimensional truncated multi-variate normal distribution I: fundamentals
复制标题
计算 N 维截断多元正态分布期望的 O(N) 算法 I:基础知识
DOI:
10.1007/s10444-021-09888-1
复制
发表时间:
2021
影响因子:
1.7
通讯作者:
Turkiyyah, George
中科院分区:
文献类型:
--
作者:
Huang, Jingfang;Cao, Jian;Fang, Fuhui;Genton, Marc G.;Keyes, David E.;Turkiyyah, George
In this paper, we present the fundamentals of a hierarchical algorithm for computing theN-dimensional integralrepresenting the expectation of a functionH(X) wheref(x|A) is the truncated multi-variate normal (TMVN) distribution with zero mean,xis the vector of integration variables for theN-dimensional random vectorX,Ais the inverse of the covariance matrixΣ, andaandbare constant vectors. The algorithm assumes thatH(x) is “low-rank” and is designed for properly clusteredXso that the matrixAhas “low-rank” blocks and “low-dimensional” features. We demonstrate the divide-and-conquer idea whenAis a symmetric positive definite tridiagonal matrix and present the necessary building blocks and rigorous potential theory–based algorithm analysis whenAis given by theexponential covariance model. The algorithm overall complexity isO(N) forN-dimensional problems, with a prefactor determined by the rank of the off-diagonal matrix blocks and number of effective variables. Very high accuracy results forNas large as 2048 are obtained on a desktop computer with 16G memory using the fast Fourier transform (FFT) and non-uniform FFT to validate the analysis. The current paper focuses on the ideas using the simple yet representative examples where the off-diagonal matrix blocks are rank 1 and the number of effective variables is bounded by 2, to allow concise notations and easier explanation. In a subsequent paper, we discuss the generalization of current scheme using the sparse grid technique for higher rank problems and demonstrate how all the moments ofkthorder or less (a total ofO(Nk) integrals) can be computed usingO(Nk) operations fork≥ 2 andoperations fork= 1.
登录
查看更多内容
DOI:
--
发表时间:
2007
期刊:
影响因子:
--
作者:
Yamamoto;et al.
通讯作者:
et al.
DOI:
10.1111/rssb.12162
发表时间:
2017-01-01
影响因子:
5.8
作者:
Botev, Z. I.
通讯作者:
Botev, Z. I.
影响因子:
1.8
作者:
Ioannis Phinikettos;A. Gandy
通讯作者:
A. Gandy
DOI:
--
发表时间:
2014
期刊:
影响因子:
--
作者:
T. M. Karbach;G. Raven;M. Schiller
通讯作者:
M. Schiller
DOI:
10.1017/cbo9781139248891
发表时间:
2018-03
期刊:
--
影响因子:
--
作者:
A. Azzalini;A. Capitanio
通讯作者:
A. Azzalini;A. Capitanio