Article
Statistics & Probability
Tetsuya Kaji, Veronika Rockova
Summary: This article develops a Bayesian computational platform for posterior sampling and optimization in models with difficult-to-evaluate marginal likelihoods. By reframing the likelihood function estimation problem as a classification problem, the authors propose likelihood (ratio) estimators that can be used in the Metropolis-Hastings algorithm. Inspired by contrastive learning and Generative Adversarial Networks (GAN), the resulting Markov chains generate samples from an approximate posterior and the asymptotic properties are characterized. The article also discusses the convergence rate and provides asymptotic normality results to justify the approach's inferential potential. The usefulness of the method is illustrated with challenging examples for existing Bayesian likelihood-free approaches.
JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
(2022)
Article
Environmental Studies
D. P. Yang, T. Liu, X. M. Zhang, X. H. Zeng, D. F. Song
Summary: Due to the complexity of the current traffic environment, existing methods of constructing driving cycles have large errors and poor representativeness, making it difficult to accurately reflect the fuel consumption and emissions of vehicles on actual roads. To address this problem, a high-precision construction method based on stepwise regression characteristic parameter selection is proposed, combining the advantages of Genetic Algorithm (GA) and Metropolis-Hastings Sampling (MHS) methods. Experimental verification and algorithm comparison demonstrate that the driving cycle constructed using the MHS-GA method is more representative, with small relative deviations in characteristic parameters and fuel consumption.
TRANSPORTATION RESEARCH PART D-TRANSPORT AND ENVIRONMENT
(2023)
Article
Environmental Sciences
Zhiyi Wang, Yoshiki Nishi
Summary: This paper proposes a stochastic model to simulate the occurrence and levels of PCBs in juvenile tuna, considering both the transport of PCBs in the ocean and biomagnification in fish. The model accommodates the uncertainty in fish's exposure to PCBs by adopting a random sampling approach. The simulation results align well with previous studies and demonstrate the model's sensitivity to the spatial distribution patterns of PCBs and moderate sensitivity to current velocity. The model holds potential for extension to more realistic scenarios and for serving as an environmental risk assessment tool.
ECOTOXICOLOGY AND ENVIRONMENTAL SAFETY
(2022)
Article
Quantum Science & Technology
Hui-Min Li, Jin-Min Liang, Zhi-Xi Wang, Shao-Ming Fei
Summary: The Quantum Approximate Optimization Algorithm (QAOA) is a promising variational quantum algorithm for combinatorial optimization problems but is limited by its requirement on the mapping to Ising Hamiltonians and nonconvex optimization landscapes. This paper presents a general method to obtain the Ising Hamiltonians for constrained combinatorial optimization problems (CCOPs) and introduces the Metropolis-Hastings warm-starting algorithm for QAOA, which guarantees convergence to global optimal solutions. The effectiveness of this method is demonstrated on the minimum weight vertex cover (MWVC), minimum vertex cover (MVC), and maximal independent set problems. The Ising Hamiltonian for the MWVC problem is obtained for the first time using this method, and the advantages of the Metropolis-Hastings warm-starting algorithm are numerically analyzed.
ADVANCED QUANTUM TECHNOLOGIES
(2023)
Article
Biology
Jure Vogrinc, Samuel Livingstone, Giacomo Zanella
Summary: We study a class of first-order locally balanced Metropolis-Hastings algorithms and examine how to optimize algorithm efficiency by choosing balancing functions and noise distributions. The research reveals that all members of the class have similar optimal acceptance rates and scaling laws as the dimension tends to infinity under certain smoothness assumptions and when the target distribution is of product form. Numerical simulations confirm the theoretical findings and show that a bimodal noise distribution in the Barker proposal yields a more efficient algorithm.
Article
Management
Sareh Nabi, Houssam Nassif, Joseph Hong, Hamed Mamani, Guido Imbens
Summary: Adding domain knowledge as a prior in learning systems has been shown to improve results. This study proposes a hierarchical empirical Bayes approach that addresses the challenges of lacking informative priors and controlling parameter learning rates. By learning empirical meta-priors and decoupling learning rates of different feature groups, the method improves performance and convergence time.
MANAGEMENT SCIENCE
(2022)
Article
Computer Science, Theory & Methods
John Moriarty, Jure Vogrinc, Alessandro Zocca
Summary: The study aims to enhance the exploration of the general-purpose random walk Metropolis algorithm in the presence of non-convex support targets. By reusing rejected proposals in A(c), the performance is improved. The algorithm falls under the Metropolis class and demonstrates consistent behavior under standard conditions. Numerical examples support the theoretical evidence of enhanced performance compared to the random walk Metropolis algorithm. Discussions on implementation issues and applications to global optimization and rare event sampling are also provided.
STATISTICS AND COMPUTING
(2021)
Article
Engineering, Civil
Y. Yang, Y. Ling, X. K. Tan, S. Wang, R. Q. Wang
Summary: This paper proposes a new method for identifying local damages in frame structures using the approximate Metropolis-Hastings (AMH) algorithm and statistical moment. The fusion index of fourth-order displacement moment and eighth-order acceleration moment is selected as the optimal damage indicator. The proposed method, which incorporates Gibbs sampling, is shown to be more time-saving and accurate in assessing damage severity compared to other similar methods. Experimental studies validate the effectiveness of the proposed method.
INTERNATIONAL JOURNAL OF STRUCTURAL STABILITY AND DYNAMICS
(2022)
Article
Mathematics, Applied
Tahani A. Abushal
Summary: This paper investigates the estimation of Akash distribution parameters when the lifetime of the product follows Type-II censoring, studying maximum likelihood estimators and Bayesian inference procedures. Approximate confidence intervals and highest posterior density intervals are derived, and the methods are compared through a Monte Carlo simulation study. The application to real data is also analyzed.
Article
Ergonomics
Jacob Mathew, Rahim F. Benekohal
Summary: The new FRA model uses a single equation for all three warning devices, while the ZINEBS model gives three different equations depending on the type of warning device used, showing closer agreement with field data. The ZINEBS model complements the new FRA model by addressing the shortcomings of the single equation.
JOURNAL OF SAFETY RESEARCH
(2021)
Article
Engineering, Civil
Jin Luo, Minshui Huang, Chunyan Xiang, Yongzhi Lei
Summary: In this paper, a simplified population Metropolis-Hastings algorithm (SP-MH) was proposed for damage identification. The algorithm exchanges information and uses a tuning-free strategy in a relatively small population, and its effectiveness and feasibility were verified.
INTERNATIONAL JOURNAL OF STRUCTURAL STABILITY AND DYNAMICS
(2023)
Article
Engineering, Civil
Guanjun Lei, Junxian Yin, Wenchuan Wang, Hao Wang, Changshun Liu
Summary: The consistency of hydrological sequences is affected by climate change and human activities, leading to significant uncertainty in hydrological frequency analysis results. The study uses the Mann-Kendall test and Hurst coefficient method to identify and test the trend of hydrological series. The empirical mode decomposition is used to obtain the trend component and perform consistency correction, while a Bayesian model with Metropolis-Hastings sampling is constructed to estimate parameters and analyze result uncertainty.
JOURNAL OF HYDROLOGIC ENGINEERING
(2023)
Article
Radiology, Nuclear Medicine & Medical Imaging
Philip Trapp, Joscha Maier, Markus Susenburger, Stefan Sawall, Marc Kachelriess
Summary: This study proposes a simple and effective postprocessing software-based correction method of scatter artifacts in cone beam CT (CBCT) scans without specific prior knowledge. The proposed method, called empirical scatter correction (ESC), generates scatter-like basis images from each projection image by convolution operations, and subtracts a linear combination of these basis images from the original projection image. ESC can improve the image quality of CBCT scans and reduce scatter artifacts in both simulated and measured data.
Article
Automation & Control Systems
Trambak Banerjee, Qiang Liu, Gourab Mukherjee, Wenguang Sun
Summary: The NEB framework is a nonparametric empirical Bayes method for estimation in the discrete linear exponential family, with strong asymptotic properties and flexibility to incorporate various constraints, providing a unified approach to estimation. Comprehensive simulation studies and analysis of real data examples demonstrate the superiority of the NEB estimator over competing methods.
JOURNAL OF MACHINE LEARNING RESEARCH
(2021)
Article
Biochemical Research Methods
Yi Yang, Xingjie Shi, Wei Liu, Qiuzhong Zhou, Mai Chan Lau, Jeffrey Chun Tatt Lim, Lei Sun, Cedric Chuan Young Ng, Joe Yeong, Jin Liu
Summary: Spatial transcriptomics is a powerful technique for analyzing gene expression and spatial information in tissues. This study presents a method called SC-MEB for identifying cell clusters based on spatial information, and demonstrates its superiority through simulations and real data analysis.
BRIEFINGS IN BIOINFORMATICS
(2022)
