Gaussian Variational Approximations for High-dimensional State Space Models

被引:9
作者
Quiroz, Matias [1 ,2 ]
Nott, David J. [3 ,4 ]
Kohn, Robert [2 ,5 ]
机构
[1] Stockholm Univ, Dept Stat, S-10691 Stockholm, Sweden
[2] ARC Ctr Excellence Math & Stat Frontiers ACEMS, Melbourne, Vic 3010, Australia
[3] Natl Univ Singapore, Dept Stat & Data Sci, Singapore 117546, Singapore
[4] Natl Univ Singapore, Inst Operat Res & Analyt, 21 Lower Kent Ridge Rd, Singapore 119077, Singapore
[5] Univ New South Wales, UNSW Business Sch, Sch Econ, Sydney, NSW 2052, Australia
来源
BAYESIAN ANALYSIS | 2023年 / 18卷 / 03期
基金
澳大利亚研究理事会;
关键词
dynamic factor; stochastic gradient; spatio-temporal modelling; MULTIVARIATE STOCHASTIC VOLATILITY; INFERENCE; ERROR; BAYES;
D O I
10.1214/22-BA1332
中图分类号
O1 [数学];
学科分类号
0701 ; 070101 ;
摘要
We consider a Gaussian variational approximation of the posterior density in high-dimensional state space models. The number of parameters in the covariance matrix of the variational approximation grows as the square of the number of model parameters, so it is necessary to find simple yet effective parametrisations of the covariance structure when the number of model parame-ters is large. We approximate the joint posterior density of the state vectors by a dynamic factor model, having Markovian time dependence and a factor covariance structure for the states. This gives a reduced description of the dependence struc-ture for the states, as well as a temporal conditional independence structure sim-ilar to that in the true posterior. We illustrate the methodology on two examples. The first is a spatio-temporal model for the spread of the Eurasian collared-dove across North America. Our approach compares favorably to a recently proposed ensemble Kalman filter method for approximate inference in high-dimensional hi-erarchical spatio-temporal models. Our second example is a Wishart-based multi-variate stochastic volatility model for financial returns, which is outside the class of models the ensemble Kalman filter method can handle.
引用
收藏
页码:989 / 1016
页数:28
相关论文
共 50 条
[31]   Permutation testing in high-dimensional linear models: an empirical investigation [J].
Hemerik, Jesse ;
Thoresen, Magne ;
Finos, Livio .
JOURNAL OF STATISTICAL COMPUTATION AND SIMULATION, 2021, 91 (05) :897-914
[32]   In Nonparametric and High-Dimensional Models, Bayesian Ignorability is an Informative Prior [J].
Linero, Antonio R. .
JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION, 2024, 119 (548) :2785-2798
[33]   High-dimensional factor copula models with estimation of latent variables [J].
Fan, Xinyao ;
Joe, Harry .
JOURNAL OF MULTIVARIATE ANALYSIS, 2024, 201
[34]   Scalable Algorithms for Learning High-Dimensional Linear Mixed Models [J].
Tan, Zilong ;
Roche, Kimberly ;
Zhou, Xiang ;
Mukherjee, Sayan .
UNCERTAINTY IN ARTIFICIAL INTELLIGENCE, 2018, :259-268
[35]   Sparse signal shrinkage and outlier detection in high-dimensional quantile regression with variational Bayes [J].
Lim, Daeyoung ;
Park, Beomjo ;
Nott, David ;
Wang, Xueou ;
Choi, Taeryon .
STATISTICS AND ITS INTERFACE, 2020, 13 (02) :237-249
[36]   On Epistemics in Expected Free Energy for Linear Gaussian State Space Models [J].
Koudahl, Magnus T. ;
Kouw, Wouter M. ;
de Vries, Bert .
ENTROPY, 2021, 23 (12)
[37]   Copula Variational LSTM for High-Dimensional Cross-Market Multivariate Dependence Modeling [J].
Xu, Jia ;
Cao, Longbing .
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, 2023, :16233-16247
[38]   Contextual anomaly detection for high-dimensional data using Dirichlet process variational autoencoder [J].
Kim, Hyojoong ;
Kim, Heeyoung .
IISE TRANSACTIONS, 2023, 55 (05) :433-444
[39]   Adaptive penalization in high-dimensional regression and classification with external covariates using variational Bayes [J].
Velten, Britta ;
Huber, Wolfgang .
BIOSTATISTICS, 2021, 22 (02) :348-364
[40]   Fitting Structural Equation Models via Variational Approximations [J].
Dang, Khue-Dung ;
Maestrini, Luca .
STRUCTURAL EQUATION MODELING-A MULTIDISCIPLINARY JOURNAL, 2022, 29 (06) :839-853