Article
Automation & Control Systems
Sumit Mukherjee, Subhabrata Sen
Summary: We study high-dimensional bayesian linear regression with product priors. We derive sufficient conditions for the leading-order correctness of the naive mean-field approximation to the log-normalizing constant of the posterior distribution using the theory of non-linear large deviations. Assuming a true linear model for the observed data, we derive a limiting infinite dimensional variational formula for the log normalizing constant. Additionally, we establish a unique optimizer for the variational problem under an additional separation condition, which governs the probabilistic properties of the posterior distribution. We provide intuitive sufficient conditions for the validity of this separation condition and illustrate the results using concrete examples.
JOURNAL OF MACHINE LEARNING RESEARCH
(2022)
Article
Biochemical Research Methods
Qian Gao, Yu Zhang, Hongwei Sun, Tong Wang
Summary: This paper reviews the methods for estimating causal effects in observational studies and evaluates their performance in high-dimensional settings. The simulation experiments show that GLiDeR and hdCBPS approaches perform well in terms of estimation accuracy, but further studies are needed for constructing valid confidence intervals.
BRIEFINGS IN BIOINFORMATICS
(2022)
Article
Statistics & Probability
Elizabeth L. Ogburn, Oleg Sofrygin, Ivan Diaz, Mark J. van der Laan
Summary: This study focuses on semiparametric estimation and inference for causal effects using observational data from a single social network. The authors propose new methods that allow for dependence among observations, considering both information transmission across network ties and latent similarities among nodes. The study also reanalyzes a controversial study on obesity causal peer effects using social network data from the Framingham Heart Study, finding no evidence for such effects after accounting for network structure.
JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
(2022)
Article
Statistics & Probability
Emanule Aliverti, Massimiliano Russo
Summary: Recent interest in Bayesian modeling of high-dimensional networks using latent space approaches has led to the development of scalable algorithms for conducting approximate Bayesian inference via stochastic optimization. These algorithms leverage sparse representations of network data to provide massive computational and storage benefits, allowing for inference in settings with thousands of nodes. An R package with efficient c++ implementation of the algorithms is provided for practical use.
JOURNAL OF COMPUTATIONAL AND GRAPHICAL STATISTICS
(2022)
Article
Computer Science, Artificial Intelligence
Kai Lagemann, Christian Lagemann, Bernd Taschler, Sach Mukherjee
Summary: Causal learning is a key challenge in scientific artificial intelligence, and the proposed deep neural architecture effectively learns causal relationships between variables from high-dimensional data and prior knowledge. The combination of convolutional and graph neural networks allows for the identification of novel causal relationships under conditions of high dimensionality, noise, and data limitations.
NATURE MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Interdisciplinary Applications
Xizhen Cai, Yeying Zhu, Yuan Huang, Debashis Ghosh
Summary: Causal mediation analysis aims to study the direct effects of exposure on outcome and the mediated effects on the pathway from exposure to outcome. This paper proposes a set of generalized structural equations to estimate the direct and indirect effects in high-dimensionality mediation analysis. Additionally, the performance of Sobel's method in obtaining standard error and confidence interval for the estimated joint indirect effect is evaluated.
COMPUTATIONAL STATISTICS & DATA ANALYSIS
(2022)
Article
Economics
Michael P. Leung
Summary: This paper examines causal inference in randomized experiments when there is network interference. Traditional models of interference assume that treatments assigned to individuals beyond a certain network distance from the ego have no effect on the ego's response. However, this assumption is violated in common models of social interactions. The researchers propose a weaker model of approximate neighborhood interference (ANI), which allows for smaller but potentially nonzero effects of treatments assigned to individuals further from the ego. Under certain conditions, they show that standard inverse-probability weighting estimators can consistently estimate useful exposure effects and have approximate normal distributions. They also consider a network heteroskedasticity and autocorrelation (HAC) variance estimator for inference.
Article
Public, Environmental & Occupational Health
Anne-Louise Ponsonby
Summary: This commentary discusses the importance of non-causal confounding in considering causal questions within a moderate-sized high-dimensional study, emphasizing the systematic evaluation of alternative explanations for exposure-disease associations to achieve the highest level of causal inference possible.
INTERNATIONAL JOURNAL OF EPIDEMIOLOGY
(2021)
Article
Mathematics
Jieyi Yi, Niansheng Tang
Summary: In this paper, a variational Bayesian approach is proposed to simultaneously select variables and estimate parameters in high-dimensional linear mixed models. Compared to traditional methods, this approach is more efficient and flexible. Simulation studies and an empirical analysis demonstrate the performance and practicality of the proposed method.
Article
Statistics & Probability
Weichang Yu, Sara Wade, Howard D. Bondell, Lamiae Azizi
Summary: High-dimensional classification and feature selection tasks are common with the advancement of data acquisition technology. In fields such as biology, genomics, and proteomics, where data are often functional and exhibit roughness and nonstationarity, traditional methods face additional challenges. In this work, we propose a novel approach called Gaussian process discriminant analysis (GPDA) that combines variable selection and classification in a unified framework. By utilizing sparse inverse covariance matrices and variational methods, our approach achieves scalable inference and demonstrates good performance.
JOURNAL OF COMPUTATIONAL AND GRAPHICAL STATISTICS
(2023)
Article
Computer Science, Artificial Intelligence
Yan Zeng, Zhifeng Hao, Ruichu Cai, Feng Xie, Libo Huang, Shohei Shimizu
Summary: This article proposes a new causal model (HDDM) to identify the causal orderings among multiple high-dimensional variables. The method derives two candidate selection rules to mitigate the ordering errors caused by violated assumptions. The efficacy of the proposed method is demonstrated through simulations on synthetic and real-world data, where the robustness of the method to noise is also verified.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Construction & Building Technology
Niloofar Moosavi, Jenny Haeggstroem, Xavier de Luna
Summary: Important advancements have been made in developing procedures for uniformly valid inference for low dimensional causal parameters in the presence of high-dimensional nuisance models. This paper reviews the literature on uniformly valid causal inference and discusses the trade-offs associated with using uniformly valid inference procedures. The study proposes estimators that uniformly converge in distribution over a class of data generating mechanisms in order to address the problem of badly approximated finite sample distributions of naive estimation strategies.
ENERGY AND BUILDINGS
(2023)
Article
Statistics & Probability
Likai Chen, Weining Wang, Wei Biao Wu
Summary: This paper introduces a method for detecting multiple change-points in high-dimensional time series, supported by theoretical consistency and asymptotic distribution. The proposed two-step procedure can capture both the biggest break across different coordinates and aggregating simultaneous breaks over multiple coordinates.
JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
(2022)
Article
Computer Science, Information Systems
Qianqian Xie, Prayag Tiwari, Deepak Gupta, Jimin Huang, Min Peng
Summary: This paper proposes a semantic reinforcement neural variational sparse topic model (SR-NSTM), which models the generative process of texts with probabilistic distributions parameterized by neural networks and achieves interpretable and sparse latent text representation learning.
INFORMATION PROCESSING & MANAGEMENT
(2021)
Article
Engineering, Multidisciplinary
Liang Yan, Tao Zhou
Summary: Bayesian computation is crucial in modern machine learning and statistics for dealing with uncertainty. Stein variational gradient descent (SVGD) is an important approximate inference algorithm, but it requires calculating the gradient of the target density. This paper proposes using a local surrogate to address this challenge, with a focus on refining the surrogate approximation using a deep neural network.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)