Article
Computer Science, Information Systems
Fujiao Ju, Yanfeng Sun, Jianqiang Li, Yaxiao Zhang, Xinglin Piao
Summary: Dimensionality reduction plays a crucial role in image recognition and data mining. Traditional methods extract features from data without considering the structure information, while this paper proposes a new approach that combines graph embedding and principal component analysis (PCA) to extract more effective features by capturing the complex relationships among data. Experimental results on publicly available datasets demonstrate that the proposed model outperforms existing classical algorithms in terms of classification accuracy.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Jingyu Wang, Lin Wang, Feiping Nie, Xuelong Li
Summary: Feature selection and feature extraction are two main strategies in data dimensionality reduction, each with its own advantages and disadvantages. This article proposes a method that combines both strategies by adding constraints and using a new algorithm for sparse projection, and introduces a "purification matrix" to eliminate meaningless information, demonstrating the effectiveness of the method for data dimensionality reduction.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Xiaofeng Wang, Leyan Shi, Jun Liu, Minglu Zhang
Summary: This paper introduces a novel cosine 2DPCA method to tackle the issues in two-dimensional principal component analysis. By applying the cosine objective function and a greedy iterative algorithm, this method demonstrates significant improvements in reconstruction, correlation, complexity, and classification.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Thermodynamics
Anirudh Jonnalagadda, Shubham Kulkarni, Akash Rodhiya, Hemanth Kolla, Konduri Aditya
Summary: Principal Component Analysis (PCA) is a widely used dimensionality reduction technique in combustion simulations. However, it may fail to capture important localized chemical dynamics. In this paper, we propose a new method called co-kurtosis PCA (CoK-PCA) that can better identify directions representing stiff dynamics. We validate the potential of CoK-PCA using synthetically generated data and compare its accuracy with PCA for real combustion datasets.
COMBUSTION AND FLAME
(2023)
Article
Computer Science, Information Systems
Huanxing Zhang, Hongxu Bi, Xiaofeng Wang, Peng Zhang
Summary: This paper proposes a new method for two-dimensional principal component analysis called 2DPCA-2-Lp. By combining the 2-norm and l(p)-norm metrics in the objective function, this method aims to maximize the ratio of projected vectors to image row vectors and achieve the dual objectives of directly maximizing projection distances and indirectly minimizing reconstruction errors. Experimental results show that 2DPCA-2-Lp outperforms most existing 2DPCA methods in terms of reconstruction and classification performances and has better robustness against noise.
INFORMATION SCIENCES
(2023)
Article
Computer Science, Artificial Intelligence
Chaofan Hu, Shuilong He, Yanxue Wang
Summary: A novel classification technique utilizing kernelled support tensor machine (KSTM) and multilinear principal component analysis (MPCA) is introduced for fault detection of rotating machines. Through the use of a 3-way tensor, KSTM, and MPCA, the technique effectively handles fault diagnosis and information classification in rotating machinery.
APPLIED INTELLIGENCE
(2021)
Article
Computer Science, Information Systems
Hao Zheng, Liyong Fu, Qiaolin Ye
Summary: Robust principal component analysis (PCA) has been proven effective in data reconstruction and recognition tasks. However, existing methods often suffer from performance and robustness issues. To address this, we propose a new method called flexible capped PCA (FCPCA) that uses capped L2,p-norm distance metric to minimize reconstruction errors. Experimental results demonstrate that FCPCA outperforms existing methods in terms of power and flexibility.
INFORMATION SCIENCES
(2022)
Article
Computer Science, Artificial Intelligence
Jingwei Chen, Jianyong Zhu, Hongyun Jiang, Hui Yang, Feiping Nie
Summary: This article proposes a clustering method called P_SFCM that combines principal component analysis and membership learning to improve the robustness of noise. Experimental results show that P_SFCM is competitive with other methods.
IEEE TRANSACTIONS ON FUZZY SYSTEMS
(2023)
Article
Geochemistry & Geophysics
Tian Zhang, Jun Wang, Erlei Zhang, Kai Yu, Yongqin Zhang, Jinye Peng
Summary: This study proposes a random multiscale convolutional network (RMCNet) incorporating a multiscale dimensionality reduction module (MDRM) to improve the accuracy of spectral-spatial classification of hyperspectral images. Experimental results demonstrate that RMCNet can deliver competitive performance compared to state-of-the-art methods.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2021)
Article
Computer Science, Artificial Intelligence
Tianyue Zhang, Furao Shen, Tao Zhu, Jian Zhao
Summary: This article proposes a fast and robust incremental subspace learning framework called EOCA for dimensionality reduction. EOCA determines the target dimensionality automatically by adaptive thresholds and extracts the low-dimensional representation of data in the form of orthogonal subspace bases. Additionally, EOCA provides the ability to merge subspaces and eliminate outlier effects.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Computer Science, Information Systems
Xiangfei Yang, Wensi Wang, Liming Liu, Yuanhai Shao, Liting Zhang, Naiyang Deng
Summary: 2DPCA based on Tl-1 criterion is proposed in this paper, which is more robust to noise than classical 2DPCA, and experimental results show its superior performance.
Article
Computer Science, Artificial Intelligence
K. Aditya Shastry, H. A. Sanjay
Summary: Data pre-processing is a technique that transforms raw data into a useful format for machine learning, with feature selection and feature extraction being significant components. This study proposes a hybrid strategy using modified Genetic Algorithm and weighted Principal Component Analysis for selecting and extracting features from agricultural datasets, resulting in significant improvements in benchmark and real-world farming datasets.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Engineering, Electrical & Electronic
Yongshan Zhang, Yang Wang, Xiaohong Chen, Xinwei Jiang, Yicong Zhou
Summary: This paper proposes a dual graph autoencoder (DGAE) for learning discriminative representations of hyperspectral images (HSIs). DGAE utilizes spatial information and band correlations to construct similarity graphs and extract feature representations. Experimental results show the superiority of DGAE over existing methods.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2022)
Article
Engineering, Civil
Anand Kumar Agrawal, Goutam Chakraborty
Summary: This paper proposes the use of neighborhood component analysis, a new dimensionality reduction algorithm, to address the issue of high dimensional feature vectors in structural health monitoring. Compared to the traditional Principal Component Analysis, neighborhood component analysis takes into account the class labels of initial feature vectors, resulting in improved classification performance in damage classification problems.
Article
Computer Science, Information Systems
Xiaoqian Zhang, Zhen Tan, Huaijiang Sun, Zungang Wang, Mingwei Qin
Summary: This paper proposes a robust image feature extraction model based on Orthogonal Low-Rank Projection Learning (OLRPL), which introduces an orthogonal matrix to preserve the main components of the sample. The row sparsity constraint is applied on the projection matrix to encourage compact, discriminative, and interpretable features. Weighted Truncated Schatten p-norm (WTSN) is proposed to solve the optimization problem of low-rank constraints, and correntropy is used to suppress complex noise in the data. Experimental results show that OLRPL performs better than existing advanced methods in terms of effectiveness and robustness.
IEEE TRANSACTIONS ON MULTIMEDIA
(2022)
Article
Engineering, Biomedical
Xiaofeng Xie, Zhu Liang Yu, Haiping Lu, Zhenghui Gu, Yuanqing Li
IEEE TRANSACTIONS ON NEURAL SYSTEMS AND REHABILITATION ENGINEERING
(2017)
Article
Computer Science, Artificial Intelligence
Qiquan Shi, Haiping Lu, Yiu-Ming Cheung
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2018)
Article
Computer Science, Artificial Intelligence
Peizhen Bai, Yan Ge, Fangling Liu, Haiping Lu
PATTERN RECOGNITION
(2019)
Article
Computer Science, Artificial Intelligence
Qiquan Shi, Yiu-Ming Cheung, Qibin Zhao, Haiping Lu
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2019)
Article
Automation & Control Systems
Yang Zhou, Haiping Lu, Yiu-Ming Cheung
Summary: Subspace learning for tensors has gained increasing interest, leading to the development of multilinear extensions of PCA and PPCA. Existing methods like Tucker-based multilinear PPCAs face issues like rotational ambiguity, while CP-based ones are more prone to overfitting. To address these problems, a new CP-based method PROTA is proposed, along with concurrent regularization strategies to alleviate overfitting.
IEEE TRANSACTIONS ON CYBERNETICS
(2021)
Article
Engineering, Electrical & Electronic
Liyan Song, Shuo Zhou, Haiping Lu
Summary: Independent Component Analysis (ICA) is a fundamental method for Blind Source Separation (BSS). This paper introduces a new approach called RAndom Matrix ICA (RAMICA), which works directly with third-order data tensors and preserves more structural information and general BSS assumptions. The RAMICA model and algorithm are developed by defining new statistics for random matrices and new procedures for whitening and independent component estimation. Experimental results demonstrate the superior BSS performance of RAMICA compared to competing methods.
Article
Computer Science, Artificial Intelligence
Hao Xu, Shengqi Sang, Peizhen Bai, Ruike Li, Laurence Yang, Haiping Lu
Summary: This paper proposes a flexible and efficient GripNet framework for graph representation learning. By introducing a new supergraph data structure and utilizing multi-layer information propagation, GripNet learns low-dimensional vector representations for different types of entities and relations, outperforming competing methods in link prediction, node classification, and data integration tasks.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Artificial Intelligence
Li Zhang, Heda Song, Nikolaos Aletras, Haiping Lu
Summary: Graph convolutional network (GCN) is an effective neural network model for graph representation learning. This paper proposes a new node-feature convolutional (NFC) layer to tackle the limitations of standard GCN. Experimental results show that NFC-GCN outperforms state-of-the-art methods in node classification.
PATTERN RECOGNITION
(2022)
Article
Computer Science, Interdisciplinary Applications
Mwiza Kunda, Shuo Zhou, Gaolang Gong, Haiping Lu
Summary: This paper proposes new methods for multi-site autism classification using the ABIDE dataset. By introducing a new measure of functional connectivity and adopting a domain adaptation approach, the classification accuracy can be improved to 73%.
IEEE TRANSACTIONS ON MEDICAL IMAGING
(2023)
Article
Computer Science, Interdisciplinary Applications
Lawrence Schobs, Andrew J. Swift, Haiping Lu
Summary: Automatic anatomical landmark localization has improved with deep learning methods. Quantifying the uncertainty of predictions is essential for clinical settings. We propose Quantile Binning to categorize predictions by uncertainty. Our framework can be applied to any continuous uncertainty measure, allowing identification of predictions with error bounds. We compare multiple uncertainty measures and demonstrate the effectiveness of Quantile Binning on landmarks with high aleatoric uncertainty.
IEEE TRANSACTIONS ON MEDICAL IMAGING
(2023)
Article
Computer Science, Artificial Intelligence
Peizhen Bai, Filip Miljkovic, Bino John, Haiping Lu
Summary: DrugBAN is a deep bilinear attention network framework that explicitly learns the local interactions between drugs and targets, and adapts to out-of-distribution data. It achieves the best performance on three benchmark datasets compared to five state-of-the-art baseline models. The visualized bilinear attention map provides interpretable insights from prediction results.
NATURE MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Annamaria Carusi, Peter D. D. Winter, Iain Armstrong, Fabio Ciravegna, David G. G. Kiely, Allan Lawrie, Haiping Lu, Ian Sabroe, Andy Swift
Summary: Despite low acceptance of medical AI applications in real-world clinical settings, collaborative process involving experts from different fields is crucial in tackling transparency issues and building trust.
NATURE MACHINE INTELLIGENCE
(2023)
Proceedings Paper
Computer Science, Information Systems
Haiping Lu, Xianyuan Liu, Shuo Zhou, Robert Turner, Peizhen Bai, Raivo E. Koot, Mustafa Chasmai, Lawrence Schobs, Hao Xu
Summary: PyKale is a Python library for knowledge-aware machine learning from multiple sources of data, aiming to accelerate interdisciplinary research. It provides a standardized six-step pipeline and focuses on leveraging knowledge from multiple sources for accurate and interpretable predictions. PyKale also includes interdisciplinary examples in bioinformatics, knowledge graph, image/video recognition, and medical imaging.
PROCEEDINGS OF THE 31ST ACM INTERNATIONAL CONFERENCE ON INFORMATION AND KNOWLEDGE MANAGEMENT, CIKM 2022
(2022)
Proceedings Paper
Astronomy & Astrophysics
Jessica R. Zheng, Vladimir Churilov, Robert Content, Jon Lawrence, Bozhong Gu, Haiping Lu, Haikun Wen, Xiangyan Yuan
GROUND-BASED AND AIRBORNE INSTRUMENTATION FOR ASTRONOMY VII
(2018)
Proceedings Paper
Computer Science, Information Systems
Qiquan Shi, Haiping Lu, Yiu-ming Cheung
CIKM'17: PROCEEDINGS OF THE 2017 ACM CONFERENCE ON INFORMATION AND KNOWLEDGE MANAGEMENT
(2017)