Article
Computer Science, Artificial Intelligence
Danyang Wu, Feiping Nie, Jitao Lu, Rong Wang, Xuelong Li
Summary: This paper proposes a structured graph learning (SGL) framework to improve clustering stability. SGL adaptively learns a structured affinity graph that contains exact k connected components, allowing clustering assignments to be directly obtained based on the graph connectivity. The simultaneous construction of the graph and structured graph learning effectively alleviates the mismatch problem, and an efficient algorithm is proposed to solve the optimization problems involved. Numerical experiments on synthetic and real datasets demonstrate the effectiveness of the proposed methods.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Statistics & Probability
Rebecca Morrison, Ricardo Baptista, Estelle Basor
Summary: This paper investigates the relationship between the covariance and precision matrices and independence properties in a certain class of non-Gaussian distributions. It proves that the correspondence holds exactly for the covariance and approximately for the precision. Analytic and numerical examples are provided to demonstrate these results.
JOURNAL OF MULTIVARIATE ANALYSIS
(2022)
Article
Mathematics
Charles Johnson, Yulin Zhang, Frank Qiu, Carla Ferreira
Summary: We investigate the converse of the known fact that if the Gershgorin discs of a real n-by-n matrix may be separated by positive diagonal similarity, then the eigenvalues are real. In the 2-by-2 case, with appropriate signs for the off-diagonal entries, we find that the converse is correct. In the 3-by-3 case, the converse is not generally correct, but it is frequently true empirically. In the n-by-n case, n >= 3, if all the 2-by-2 principal submatrices have inseparable discs (strongly inseparable discs'), the full matrix must have a nontrivial pair of conjugate complex eigenvalues (i.e., cannot have all real eigenvalues), and this hypothesis cannot generally be weakened.
LINEAR & MULTILINEAR ALGEBRA
(2023)
Article
Computer Science, Artificial Intelligence
M. Tanveer, M. A. Ganaie, P. N. Suganthan
Summary: Ensemble classifiers with random vector functional link networks have shown improved performance in classification problems. Two approaches were proposed in this paper to solve classification problems, which were evaluated on 33 datasets. The experimental results demonstrate that the proposed methods significantly outperform other baseline models.
APPLIED SOFT COMPUTING
(2021)
Article
Economics
Hanchao Wang, Bin Peng, Degui Li, Chenlei Leng
Summary: This paper presents an estimation method for covariance matrices with conditional sparse structure, addressing challenges such as estimating dense, large-dimensional, and varying matrices. The proposed method achieves uniform consistency and convergence rates under certain technical conditions, with numerical studies validating its finite-sample performance.
JOURNAL OF ECONOMETRICS
(2021)
Article
Computer Science, Software Engineering
Dan Zhang, Zhongke Wu, Xingce Wang, Chenlei Lv, Na Liu
Summary: This research introduces a novel method for measuring the similarity of 3D skulls and faces by utilizing harmonic wave kernel signature to describe intrinsic structure and distinguish basic shape and complex topology. Experimental results have shown the effectiveness of this method in measuring skull and face similarity, with both processes being independent yet utilizing the same framework.
Article
Agriculture, Dairy & Animal Science
Bjarke Grove Poulsen, Tage Ostersen, Bjarne Nielsen, Ole Fredslund Christensen
Summary: This study compared four methods for multibreed relationship matrices and found that models using the MF method or the GT method were generally more accurate and less biased for predicting breeding values with phenotypes from rotationally crossbred animals. The GT, MF, and SM methods were the least biased for prediction of breeding values in purebred animals.
GENETICS SELECTION EVOLUTION
(2022)
Article
Statistics & Probability
Yangchun Zhang, Jiaqi Chen, Bosen Cui, Boping Tian
Summary: The density function of the limiting spectral distribution (LSD) of sample covariance matrices is commonly used in large scale statistical inference with infinite sample size and dimension. However, explicit expressions for the density function generated by vector autoregressive moving average (VARMA) models are not available. For such models, where the sample covariance matrices do not have independence structure in columns, we propose the use of modified kernel estimators that have been proven to be consistent. A simulation study is conducted to demonstrate the performance of the estimators.
COMMUNICATIONS IN STATISTICS-THEORY AND METHODS
(2022)
Article
Meteorology & Atmospheric Sciences
T. Waterman, A. D. Bragg, G. Katul, N. Chaney
Summary: Earth system models and mesoscale models have been using increasingly complex parameterization schemes for the atmospheric boundary layer. This study evaluates existing parameterizations for the surface boundary conditions of potential temperature variance (PTV) and finds that the existing schemes are acceptable over a variety of surface conditions. However, they perform poorly over heterogeneous surfaces and rough landscapes. Attempts to improve the results using canopy structure and surface roughness characteristics did not significantly enhance the predictive power of the models. There is no strong evidence indicating that large scale circulations cause substantial deviations from textbook models.
JOURNAL OF GEOPHYSICAL RESEARCH-ATMOSPHERES
(2022)
Article
Chemistry, Multidisciplinary
Yifang Yang, Fei Li
Summary: In this paper, a novel subspace clustering method called Kernel Block Diagonal Representation Subspace Clustering with Similarity Preservation (KBDSP) is proposed. KBDSP generates an affinity matrix with a block diagonal structure by introducing a block diagonal representation term, and constructs a similarity-preserving regularizer that minimizes the discrepancy between inner products of the original data and inner products of the reconstructed data in the kernel space, thereby better preserving the similarity information between the original data. Experimental results demonstrate the effectiveness of the proposed method.
APPLIED SCIENCES-BASEL
(2023)
Article
Engineering, Electrical & Electronic
Dan Li, Fanqiang Kong, Qiang Wang
Summary: The paper introduces a novel nonlocal joint kernel sparse representation method based on local covariance (NLJKSR-LC) for hyperspectral image (HSI) classification, aiming to achieve better classification results with a small amount of training samples. By combining both global and local information, this method can effectively explore spectral-spatial features and improve the accuracy of classification results.
Article
Engineering, Electrical & Electronic
Xin Wang, Yuxi Li, Gang Hao
Summary: This article focuses on suboptimal fusion estimation weighted by matrices for multisensors linear systems with unknown cross-covariance. The constraints for positive definiteness and consistency are derived, and a suboptimal fusion estimation algorithm based on linear matrix inequality (LMI) algorithm combined with AutoML is proposed. Simulation analyses confirm the effectiveness and correctness of the conclusion.
IEEE SENSORS JOURNAL
(2023)
Article
Statistics & Probability
Jiayu Lai, Xiaoyi Wang, Kaige Zhao, Shurong Zheng
Summary: The main focus of this study is to test the structure of a covariance matrix in a high-dimensional setting, which is crucial in financial stock analyses, genetic series analyses, and other fields. We propose a test framework based on U-statistics to relax the assumptions of normality, two-diagonal block, or sub-block dimensionality. The asymptotic distributions of the U-statistics are established under the null and local alternative hypotheses. Furthermore, a test approach is developed for alternatives with different sparsity levels. Both simulation study and real data analysis demonstrate the performance of our proposed methods.
Article
Engineering, Multidisciplinary
Antonina M. Kosikova, Omid Sedehi, Costas Papadimitriou, Lambros S. Katafygiotis
Summary: This paper investigates Bayesian model updating based on Gaussian Process models by reformulating the problem and proposing a new kernel function selection method, aiming to balance fitting accuracy, generalizability, and model parsimony. Computational issues are addressed using Laplace approximation and sampling techniques, and numerical and experimental examples are provided to demonstrate the accuracy and robustness of the proposed framework.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2023)
Article
Computer Science, Information Systems
Birsen Eygi Erdogan, Sureyya Ozogur-Akyuz, Pinar Karadayi Atas
Summary: Ensemble machine learning methods, specifically using Support Vector Machines, were utilized to construct a weighted functional margin classifier ensemble in order to differentiate between solid and unhealthy banks in Turkey. A novel ensemble generation method enhanced by a pruning strategy and a novel aggregation approach using weighted sums were developed to improve prediction performance. Results demonstrated the proposed ensemble method outperformed traditional SVM and logistic regression models.
INFORMATION SCIENCES
(2021)
Article
Computer Science, Artificial Intelligence
Chaodie Liu, Feiping Nie, Rong Wang, Xuelong Li
Summary: Spectral clustering has gained increasing attention for its well-defined mathematical frameworks and superior performance. However, there are still two limitations to be solved: the unpredictable deviation in clustering results and the lack of interpretability for data points in the boundary area. To address these challenges, a graph-based soft-balanced fuzzy clustering (GBFC) model is proposed. It explicitly preserves the nonnegative property of the clustering indicator matrix and imposes row normalization on it. Additionally, a novel balanced constraint is designed to regularize the clustering results and constrain the cluster size.
IEEE TRANSACTIONS ON FUZZY SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Feiping Nie, Ziheng Li, Rong Wang, Xuelong Li
Summary: K-means is a simple and popular clustering algorithm, widely used in machine learning research. It aims to find cluster centers and minimize the squared distances between samples and their nearest centers. In this paper, a novel K-means algorithm is proposed, which reformulates the objective function and introduces an efficient iterative re-weighted algorithm for optimization. The algorithm shows faster convergence rate compared to the Lloyd's algorithm, while maintaining the same computational complexity. Experimental results on benchmark datasets demonstrate the effectiveness and efficiency of the proposed algorithm.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Zhoumin Lu, Feiping Nie, Rong Wang, Xuelong Li
Summary: Multi-view clustering aims to discover common patterns from multi-source data, utilizing the data-driven nature and larger search space of deep learning methods. However, deep learning methods lack interpretability compared to traditional methods that have more stability in optimization. This paper proposes a multi-view spectral clustering model that combines the advantages of both traditional and deep learning methods, achieving better performance in multi-view clustering and semi-supervised classification tasks while maintaining interpretability and stability.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Jianyong Zhu, Wenjie Zhao, Hui Yang, Feiping Nie
Summary: This article introduces an innovative spectral embedding method, AFSEFK, which combines the learning of anchor graph-based fuzzy spectral embedding model and fuzzy K-means. It optimizes the membership relationships and learns the local and global structures in the data simultaneously. The article also proposes a method to enhance the quality of similarity graphs and clustering performance.
IEEE TRANSACTIONS ON FUZZY SYSTEMS
(2023)
Article
Automation & Control Systems
Yong Peng, Honggang Liu, Wanzeng Kong, Feiping Nie, Bao-Liang Lu, Andrzej Cichocki
Summary: In this article, a joint EEG feature transfer and semi-supervised cross-subject emotion recognition model is proposed to enhance emotion recognition performance by optimizing the shared subspace projection matrix and target label. The spatial-frequency activation patterns of critical EEG frequency bands and brain regions in cross-subject emotion expression are quantitatively identified by analyzing the learned shared subspace.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2023)
Article
Computer Science, Artificial Intelligence
Jingjing Xue, Feiping Nie, Rong Wang, Xuelong Li
Summary: This paper presents a novel graph clustering model called Rank-$r$r Discrete Matrix Factorization (DMF-RR), which has linear time complexity and can directly obtain discrete solutions without post-processing. By constraining the factor matrices to be indicator matrices, DMF-RR captures the clustering structure and utilizes an anchor graph. Extensive experiments demonstrate the superiority of DMF-RR.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Feiping Nie, Xia Dong, Zhanxuan Hu, Rong Wang, Xuelong Li
Summary: This work proposes a parameter-free clustering model called discriminative projected clustering (DPC) for low-dimensional and discriminative projection learning and clustering. The connection between DPC and linear discriminant analysis (LDA) is theoretically analyzed, and experiments demonstrate the effectiveness and efficiency of DPC in real-world applications.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Feiping Nie, Sisi Wang, Zheng Wang, Rong Wang, Xuelong Li
Summary: Principal component analysis (PCA) is unable to overcome the influence of outliers. Existing robust PCA methods have limitations and a new discrete robust PCA algorithm is proposed to eliminate the influence of outliers completely and improve anomaly detection capability.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Wei Chang, Feiping Nie, Yijie Zhi, Rong Wang, Xuelong Li
Summary: Multitask learning is a joint learning paradigm that combines multiple related tasks to improve performance. It has been observed that different tasks share a low-dimensional common subspace. To approximate the rank minimization problem, two regularization-based models are proposed to minimize the k minimal singular values. Compared to the standard trace norm, these models provide tighter approximations and better capture the low-dimensional subspace among multiple tasks.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Danyang Wu, Feiping Nie, Jitao Lu, Rong Wang, Xuelong Li
Summary: This paper proposes a structured graph learning (SGL) framework to improve clustering stability. SGL adaptively learns a structured affinity graph that contains exact k connected components, allowing clustering assignments to be directly obtained based on the graph connectivity. The simultaneous construction of the graph and structured graph learning effectively alleviates the mismatch problem, and an efficient algorithm is proposed to solve the optimization problems involved. Numerical experiments on synthetic and real datasets demonstrate the effectiveness of the proposed methods.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Rong Wang, Penglei Wang, Danyang Wu, Zhensheng Sun, Feiping Nie, Xuelong Li
Summary: Recently, graph-based multi-view clustering (GMC) has attracted attention for achieving promising performance. However, existing SGL methods suffer from sparse graphs lacking useful information. To address this, we propose the M(2)SGL model, introducing multiple different order graphs into the SGL procedure. The model outperforms state-of-the-art benchmarks in empirical results.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Jingyu Wang, Zhenyu Ma, Feiping Nie, Xuelong Li
Summary: This paper introduces a simple and fast method called Efficient Discrete Clustering with Anchor Graph (EDCAG) to address the efficiency and accuracy issues of spectral clustering on large-scale data. The method uses sparse anchors to accelerate graph construction and designs an intraclass similarity maximization model to handle anchor graph cut problems and exploit more explicit data structures. It also employs a fast coordinate rising algorithm to optimize discrete labels of samples and anchors. Experimental results demonstrate the rapidity and competitive clustering effect of EDCAG.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Yu Guo, Yuan Sun, Zheng Wang, Feiping Nie, Fei Wang
Summary: In this article, a novel unsupervised feature selection model, DSFEL, is proposed. DSFEL includes a module for learning a block-diagonal structural sparse graph to represent the clustering structure and another module for learning a completely row-sparse projection matrix using the l(2,0)-norm constraint to select distinctive features. Experimental results on nine real-world datasets demonstrate that the proposed method outperforms existing state-of-the-art unsupervised feature selection methods.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Automation & Control Systems
Zhengxin Li, Feiping Nie, Danyang Wu, Zheng Wang, Xuelong Li
Summary: Classification is a fundamental task in data mining, but high-dimensional data can degrade classification performance. To solve this problem, dimensionality reduction, including feature extraction and feature selection, is usually adopted. In this study, we propose a supervised feature selection method that combines the discriminative power of linear discriminant analysis (LDA) with the advantages of feature selection. This method uses trace ratio LDA with p-norm regularization and imposes an orthogonal constraint on the projection matrix. Experimental results on synthetic and real-world datasets demonstrate the effectiveness of the proposed method.
IEEE TRANSACTIONS ON CYBERNETICS
(2023)
Article
Geochemistry & Geophysics
Xiaojun Yang, Mingjun Zhu, Bo Sun, Zheng Wang, Feiping Nie
Summary: Currently, unsupervised hyperspectral image segmentation methods face challenges due to noise and anomalies in the data. To address this, this letter proposes a method called fuzzy C-multiple-means (FCMM), which divides data points into multiple subclusters and defines c clusters. Unlike traditional clustering algorithms, FCMM uses an optimization problem for the fuzzy affiliation matrix and an alternating iterative update method to enhance robustness and reduce the effect of outliers in the datasets. Experiments on several HSI datasets validate the effectiveness of FCMM.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2023)