Article
Mathematics, Applied
Samuel H. Rudy, Themistoklis P. Sapsis
Summary: This paper considers methods for imposing sparsity in Bayesian regression and discusses the need for additional regularization or thresholding on top of automatic relevance determination (ARD). Two classes of methods, regularization-based and thresholding-based, are proposed to learn parsimonious solutions to linear problems. Analytical demonstrations show favorable performance in learning a small set of active terms in a linear system with a sparse solution.
PHYSICA D-NONLINEAR PHENOMENA
(2021)
Article
Engineering, Mechanical
Q. Dollon
Summary: Structural model updating, a branch of the discipline dedicated to Bayesian approaches, aims to match finite element models to real asset observations. This paper proposes a new forward model using a perturbed two-stage condensation technique, which improves identifiability even in the presence of sparse observations. Sensitivity analyses are performed to explore the performance and limitations of the proposed method.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Sara Hosseinzadeh Kassani, Farhood Rismanchian, Peyman Hosseinzadeh Kassani
Summary: This study combines k-nearest neighbor rule and relevance vector machines, proposing a k relevance vector model (k-RV) to optimize performance by selecting important features and considering relevancy in the feature space. Introducing a new parameter for improving classification accuracy, the k-RV model shows competitive performance in experiments compared to state-of-the-art methods.
APPLIED SOFT COMPUTING
(2021)
Article
Mathematics
Luca Martino, Fernando Llorente, Ernesto Curbelo, Javier Lopez-Santiago, Joaquin Miguez
Summary: A novel adaptive importance sampling scheme is proposed for Bayesian inversion problems, where variables of interest and data noise power are inferred using different methods. The technique involves iterative steps of sampling and optimization, with the noise power acting as a tempered parameter for the posterior distribution of the variables of interest. Numerical experiments show the benefits of the proposed approach in Bayesian analysis.
Article
Engineering, Electrical & Electronic
Moataz M. H. El Ayadi, Mahmoud H. Ismail
Summary: A novel approach for estimating parameters of the extended generalized-K (EGK) distribution is proposed based on Gibbs sampling technique. Results show that estimated and original distributions are virtually indistinguishable, and formal metrics like Kullback-Leibler (KL) divergence, mean integrated squared bias (MISB), mean integrated variance (MIV), and mean integrated squared error (MISE) all exhibit excellent agreement between the two.
Article
Statistics & Probability
Gregor Zens, Sylvia Fruehwirth-Schnatter, Helga Wagner
Summary: This paper introduces a novel latent variable representation based on Polya-Gamma random variables for Bayesian models of binary and categorical data. New Gibbs sampling algorithms and marginal data augmentation strategies are derived to improve the efficiency of Bayesian methods in this context. Extensive simulations and real data applications illustrate the merits of this approach.
JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
(2023)
Article
Ecology
Yohei Miura, Hiroomi Imamoto, Yasuhiro Asada, Masaki Sagehashi, Michihiro Akiba, Osamu Nishimura, Daisuke Sano
Summary: In this study, prediction models for algal blooms in Japanese dam reservoirs were constructed using a combination of sparse modeling algorithm and support vector machine. Relevant variables were selected using automatic relevance determination, and the models achieved high accuracy and precision in predicting the occurrence of algal blooms.
ECOLOGICAL INFORMATICS
(2023)
Article
Statistics & Probability
Niloy Biswas, Anirban Bhattacharya, Pierre E. Jacob, James E. Johndrow
Summary: This paper considers the use of Markov chain Monte Carlo (MCMC) algorithms with continuous shrinkage priors in Bayesian high-dimensional regression. The authors address the challenge of determining the number of iterations to perform, particularly in the context of modern data sets with large numbers of covariates. They propose coupling techniques tailored to this setting, allowing for practical, non-asymptotic diagnostics of convergence. By establishing conditions for drift and minorization, they prove that the proposed couplings have finite expected meeting time. Empirical results demonstrate the scalability of the proposed couplings, with less than 1000 iterations being sufficient for a Gibbs sampler to reach stationarity in a regression with 100,000 covariates.
JOURNAL OF THE ROYAL STATISTICAL SOCIETY SERIES B-STATISTICAL METHODOLOGY
(2022)
Article
Agriculture, Dairy & Animal Science
Vrinda B. Ambike, R. Venkataramanan, S. M. K. Karthickeyan, K. G. Tirumurugaan, Kaustubh G. Bhave, Jayant Khadse
Summary: Genetic parameters of semen production traits in Holstein-Friesian (HF) purebred, HF crossbred, and indigenous bulls were evaluated using Bayesian univariate and bivariate models. The study found that indigenous bulls had lower mean values for semen production traits, except for initial sperm motility. Heritability and repeatability estimates varied among different breeds and traits. There was a negative genetic correlation between ejaculate volume (EV) and sperm concentration (SC), while a positive genetic correlation was observed between initial sperm motility (ISM) and post-thaw motility (PTM). The Bayesian framework provided precise genetic parameter estimates that can contribute to the improvement of semen traits and bull fertility.
Article
Computer Science, Theory & Methods
Theodore Papamarkou
Summary: In this work, a blocked Gibbs sampling scheme is proposed to make minibatch MCMC sampling for feedforward neural networks more feasible. Partitioning the parameter space allows sampling regardless of layer width, and reducing the proposal variance in deeper layers can alleviate vanishing acceptance rates. The length of non-convergent chains can be increased to improve predictive accuracy in classification tasks, and avoiding vanishing acceptance rates enables longer chain runs with practical benefits. Non-convergent chain realizations are also useful in quantifying predictive uncertainty. However, performing minibatch MCMC sampling for feedforward neural networks in the presence of augmented data is still an open problem.
STATISTICS AND COMPUTING
(2023)
Article
Agriculture, Dairy & Animal Science
Manoj Kumar, Vikas Vohra, Poonam Ratwan, S. S. Lathwal
Summary: This study estimated the genetic parameters of milk and milk composition traits in Murrah buffaloes using data from 28 years. The results showed that all traits except for milk fat percentage had ample genetic variability that could be utilized for genetic improvement.
ANIMAL BIOTECHNOLOGY
(2022)
Article
Acoustics
Yixian Li, Xiaoyou Wang, Yong Xia, Limin Sun
Summary: In this study, a sparse Bayesian framework is proposed to simultaneously identify the force location and time history and reconstruct the responses while considering uncertainties. The approach utilizes a self-adaptive posterior maximization strategy to iteratively calculate the most probable values of unknown forces, measurement noise, and force variances. This method effectively identifies and quantifies input forces while reducing uncertainty in the full-field structural responses.
JOURNAL OF SOUND AND VIBRATION
(2023)
Article
Statistics & Probability
Roy Levy, Craig K. Enders
Summary: This study provides a coherent approach to Bayesian analysis of multilevel models in the presence of missing data, covering the main aspects of the models and missingness.
COMMUNICATIONS IN STATISTICS-SIMULATION AND COMPUTATION
(2023)
Article
Statistics & Probability
Siddharth Rawat, Soudeep Deb
Summary: This study develops a new statistical technique to capture the spatio-temporal dependence pattern of COVID-19 spread and shows that the proposed model is adequate for understanding the dependency in the data and has superior predictive power compared to other models.
JOURNAL OF APPLIED STATISTICS
(2023)
Article
Engineering, Mechanical
Quentin Dollon, Jerome Antoni, Antoine Tahan, Martin Gagnon, Christine Monette
Summary: This paper introduces a fast Gibbs sampler for solving a fully Bayesian problem in operational modal analysis. The sampler is able to infer modal properties from the FFT of well-separated modes and capture system identification and related uncertainties through a posterior distribution. The sampling scheme demonstrates fast convergence through two strategies.
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
(2022)