Article
Engineering, Mechanical
Jing Chen, Yawen Mao, Manfeng Hu, Liuxiao Guo, Quanmin Zhu
Summary: This study proposes a decomposition optimization-based expectation maximization algorithm for switching models. The identities of each sub-model are estimated in the expectation step, while the parameters are updated using the decomposition optimization method in the maximization step. Compared with the traditional expectation maximization algorithm and the gradient descent expectation maximization algorithm, the decomposition optimization-based expectation maximization algorithm avoids the matrix inversion and eigenvalue calculation; thus, it can be extended to complex nonlinear models and large-scale models. Convergence analysis and simulation examples are given to show the effectiveness of the proposed algorithm.
NONLINEAR DYNAMICS
(2023)
Article
Automation & Control Systems
Yujie Tang, Meiling Wang, Yi Yang, Ziquan Lan, Yufeng Yue
Summary: This article proposes a novel algorithm to improve the accuracy and robustness of multirobot localization by filtering out outlier data associations between local maps. The algorithm designs a mixture probability model and hierarchical expectation maximization algorithm to identify outliers and calculate the inlier probability in loop closures, thereby relieving the adverse effect of outliers on localization accuracy.
IEEE-ASME TRANSACTIONS ON MECHATRONICS
(2023)
Article
Computer Science, Artificial Intelligence
Ying Liu, Long Chen, Yunchong Li, Jun Zhao, Wei Wang
Summary: An echo state network (ESN) model with input noise is proposed to predict time series with noise. The states in the ESN are approximated by linearizing it through an extended Kalman filter (EKF) due to the introduction of input noise. The expectation maximization algorithm (EM) is used to iteratively update all uncertain parameters for model learning.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Engineering, Multidisciplinary
Artur J. Lemonte
Summary: This paper introduces a multivariate family of distributions for multivariate count data with excess zeros, deriving various general properties of the distribution, with marginal distributions being univariate zero-inflated Bell distributions. Model parameters are estimated using traditional maximum likelihood estimation methods, with a simple EM algorithm developed to calculate the maximum likelihood estimates. Empirical applications using real multivariate count data demonstrate the usefulness of the new class of distributions and offer comparisons with other related distributions.
APPLIED MATHEMATICAL MODELLING
(2022)
Article
Mathematics, Applied
Mohamed S. Eliwa, Essam A. Ahmed
Summary: This study investigates point and interval estimations of the Lomax distribution in constant stress partially accelerated life tests (ALTs) using progressive first failure type-II censored samples. The point estimates of unknown parameters and acceleration factor are obtained through maximum likelihood and Bayesian approaches. The results are reported using the Markov Chain Monte Carlo (MCMC) technique for both symmetric and asymmetric loss functions. Monte Carlo simulation studies are performed to compare results under different sample sizes. The proposed methods are then applied to oil breakdown time data in insulating fluid under two high-test voltage stress levels.
Article
Mathematics
Xiaowei Dong, Feng Sun, Fangchao Xu, Qi Zhang, Ran Zhou, Liang Zhang, Zhongwei Liang
Summary: This study effectively addressed the issue of multi-parameter estimation in the reliability analysis of electromechanical products by establishing a hybrid Weibull distribution model and designing an optimization algorithm.
Article
Automation & Control Systems
Anahita Sadeghian, Nabil Magbool Jan, Ouyang Wu, Biao Huang
Summary: This paper introduces data-driven process modeling using probabilistic principal component analysis for predicting unmeasurable or expensive-to-measure variables in the era of easily accessible data collection. It proposes a robust probabilistic principal component regression method to handle outlying observations and validates its robustness through case studies.
CHEMOMETRICS AND INTELLIGENT LABORATORY SYSTEMS
(2022)
Article
Agronomy
Liyuan Zhang, Huihui Zhang, Wenting Han, Yaxiao Niu, Jose L. Chavez, Weitong Ma
Summary: The study introduced a new crop water stress indicator, MGDEXG, and conducted experiments in a maize field in northern Colorado under varying deficit irrigation conditions. The results showed that MGDEXG was sensitive to maize water status and had significant correlations with water stress references such as CWSI.
AGRICULTURAL WATER MANAGEMENT
(2021)
Article
Computer Science, Artificial Intelligence
Florian Hirschberger, Dennis Forster, Joerg Luecke
Summary: This research presents a clustering algorithm that uses a variational approach to optimize Gaussian mixture models. The algorithm achieves efficient operation on large-scale datasets by approximating expectation maximization through truncated posteriors and coreset-based partial E-steps. The algorithm demonstrates sublinear scaling and significant speed improvements compared to existing methods.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Mathematics, Applied
Zeljko Kereta, Robert Twyman, Simon Arridge, Kris Thielemans, Bangti Jin
Summary: This study investigates two classes of algorithms for accelerating OSEM based on variance reduction, resulting in significantly improved efficiency and accuracy for penalized PET reconstructions.
Article
Chemistry, Analytical
Ossi Kaltiokallio, Roland Hostettler, Huseyin Yigitler, Mikko Valkama
Summary: This paper introduces an Expectation-Maximization (EM) algorithm based on Gaussian smoothing to estimate unknown RSS model parameters for device-free localization and tracking (DFLT) without the need for supervised training or calibration periods. The proposed system improves the accuracy of existing DFLT methods, is computationally efficient, and outperforms a state-of-the-art adaptive DFLT system in terms of tracking accuracy.
Article
Automation & Control Systems
Xinyue Wang, Junxia Ma, Weili Xiong
Summary: This paper investigates the parameter identification of bilinear state-space model (SSM) in the presence of random outliers and time-varying delays. The system output is written as a regressive form based on the observable canonical form of the bilinear model, and a bilinear state observer is used to estimate the unknown states. To eliminate the influence of outliers and time-varying delays, the Student's t distribution is employed to deal with the measurement noise, and a first-order Markov chain is used to model the delays. The unknown parameters, delays, noise variance, states, and transition probability matrix can be estimated iteratively using the expectation-maximization (EM) algorithm.
INTERNATIONAL JOURNAL OF ADAPTIVE CONTROL AND SIGNAL PROCESSING
(2023)
Article
Mathematics
Omar M. Abou Al-Ola, Ryosuke Kasai, Yusaku Yamaguchi, Takeshi Kojima, Tetsuya Yoshinaga
Summary: This article introduces a new algorithm that combines the advantages of different iterative schemes by combining ordered-subsets EM and MART with weighted geometric or hybrid means, achieving a decrease in the objective function with each iteration and outperforming OS-EM and OS-MART alone. The algorithm shows excellent performance in image reconstruction experiments, especially in dealing with noise issues.
Article
Engineering, Electrical & Electronic
Mohammed Rashid, Mort Naraghi-Pour
Summary: This paper studies the problem of downlink channel estimation in multi-user massive MIMO systems. Bayesian compressive sensing approach is employed to reduce the pilot overhead by utilizing the clustered sparse structure of the channel in the angular domain. An EP algorithm is developed to approximate the joint distribution of the sparse vector and its support, and it is used for direct estimation of the channel. EM algorithm is used to estimate the unknown model parameters. Simulation results show that the proposed EM-EP algorithm outperforms several recently-proposed algorithms.
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY
(2023)
Article
Automation & Control Systems
Manfred Jaeger
Summary: This paper presents a thorough analysis of the theoretical properties of the adaptive imputation and maximization (AIM) algorithm and its relationship with the expectation maximization (EM) algorithm. The study identifies conditions under which AIM and EM are equivalent and shows that AIM can produce consistent estimates in non-ignorable incomplete data scenarios where EM becomes inconsistent. The paper also develops a general theory of the AIM algorithm for discrete data settings and a general discretization approach for incomplete continuous data.
JOURNAL OF MACHINE LEARNING RESEARCH
(2022)