Article
Computer Science, Artificial Intelligence
Eufrasio de A. Lima Neto, Paulo C. Rodrigues
Summary: SVD is a widely used algorithm for dimensionality reduction and principal component analysis, but it is not suitable for data contaminated with outlying observations. To overcome this limitation, a kernel robust SVD algorithm is proposed, which operates in the original space and applies a robust linear regression framework to obtain robust estimates for the singular values and singular vectors. Simulation results show that the proposed algorithm outperforms classical and robust SVD algorithms. The merits of the proposed algorithm are also illustrated in an image recovery application.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Mathematics, Applied
Yonghong Duan, Ruiping Wen
Summary: This paper proposes an alternating direction power method for computing the largest singular value and singular vector of a matrix. The method is similar to the power method but requires fewer operations in the iterations. Convergence of the new method is proven under suitable conditions. Theoretical analysis and numerical experiments demonstrate that the new method is feasible and more effective than the power method in certain cases.
Article
Computer Science, Artificial Intelligence
David Vander Mijnsbrugge, Femke Ongenae, Sofie Van Hoecke
Summary: Traditionally, neural networks are represented by connected neuron layers using matrix multiplications. We propose a method that composes weight matrices using orthogonal basis matrices, which results in more parameter-efficient representations and improved performance in neural networks.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Software Engineering
Xiaotong Li, Karel Adamek, Wes Armour
Summary: This paper presents a Python-based GPU-accelerated software for singular value thresholding (SVT) in matrix completion. The software achieves high accuracy and computational efficiency, and its potential applications are demonstrated through examples of image inpainting and traffic sensing.
Article
Computer Science, Artificial Intelligence
Tianlin Huang, Rujie Zhao, Lvqing Bi, Defu Zhang, Chao Lu
Summary: The iterative nature of SVD algorithms poses a challenge that is addressed by a proposed efficient initialization method using a neural network to initialize user and item features. Experimental results show that this framework outperforms state-of-the-art methods, especially in rating prediction and item ranking.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Construction & Building Technology
Payam Parsa, Hosein Naderpour
Summary: In this study, three innovative models were proposed to estimate the peak shear strength of RC shear walls using a combination of Support Vector Regression and meta-heuristic optimization algorithms. The models were validated using a large database of experimental data, showing good accuracy in predicting the shear strength of RC shear walls. Researchers can use these models to enhance the accuracy of predicting the behavior of structures and reduce construction costs.
JOURNAL OF BUILDING ENGINEERING
(2021)
Article
Engineering, Geological
Enming Li, Fenghao Yang, Meiheng Ren, Xiliang Zhang, Jian Zhou, Manoj Khandelwal
Summary: The main purpose of blasting operations is to produce desired and optimum mean size rock fragments to improve production efficiency and reduce costs. AI-based models are popular for predicting blasting fragmentation, with the Grey Wolf Optimization Support Vector Regression model showing the best comprehensive performance.
JOURNAL OF ROCK MECHANICS AND GEOTECHNICAL ENGINEERING
(2021)
Article
Engineering, Civil
Guangyong Sun, Wenju Li, Quantian Luo, Qing Li
Summary: This study developed an effective method for identifying modal parameters of vibrating structures using DIC technique, SVD, and nonlinear least square iteration. Experimental results demonstrated the advantages of this method over other techniques, showing effective identification of modal parameters in practical applications.
THIN-WALLED STRUCTURES
(2021)
Article
Mathematics, Applied
Zhuo-Heng He, Michael K. Ng, Chao Zeng
Summary: This paper investigates the generalized singular value decompositions for two tensors via the T-product, discussing the structures of T-QSVD and T-PSVD in detail and presenting algorithms with numerical examples. For color image watermarking processing, the T-QSVD and T-PSVD approaches offer two advantages: processing two color watermarks simultaneously and only requiring one key to be saved.
NUMERICAL MATHEMATICS-THEORY METHODS AND APPLICATIONS
(2021)
Article
Geochemistry & Geophysics
Jinsheng Jiang, Wencai Yang, Haoran Ren, Mengtao Chen
Summary: Faults, fractures, karst caves, and other small-scale geological targets are important in carbonate oil and gas exploration. Traditional methods have difficulty in imaging these targets with high resolution due to their low-energy diffractions. This study proposes an improved singular value decomposition (SVD) method that effectively separates diffractions and reflections, maintaining stability even in the presence of Gaussian noise.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Computer Science, Information Systems
Michele Alessandrini, Giorgio Biagetti, Paolo Crippa, Laura Falaschetti, Lorenzo Manoni, Claudio Turchetti
Summary: The study presents an efficient implementation of the SVD algorithm on ARM Cortex-M, covering common algorithms for SVD, developing a library suitable for embedded systems without an operating system, finding the best implementation through a comparative study, and demonstrating a practical application example. The chosen algorithms have been implemented on ARM Cortex-M4F microcontrollers with limited hardware resources and experiments have been conducted to select the best algorithm performance in terms of speed, accuracy, and energy consumption.
Article
Automation & Control Systems
Weikang Wang, Chang Chen, Wenxuan Yao, Kaiqi Sun, Wei Qiu, Yilu Liu
Summary: This article proposes a novel model combining cross entropy and singular value decomposition, which can compress synchrophasor data to an extremely small size while maintaining superior accuracy and retaining critical information under disturbance conditions.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2021)
Article
Energy & Fuels
Paulino Jose Garcia Nieto, Esperanza Garcia-Gonzalo, Beatriz M. Paredes-Sanchez, Jose P. Paredes-Sanchez
Summary: This study developed an artificial smart model based on support vector machines and grid search optimizer for predicting and characterizing the Higher Heating Value (HHV) of raw biomass. The results showed that the model was accurate in predicting the HHV of biomass and highlighted the importance of physico-chemical parameters in determining the HHV.
Article
Computer Science, Interdisciplinary Applications
K. Parand, M. Razzaghi, R. Sahleh, M. Jani
Summary: In this paper, a numerical method based on least squares support vector regression is proposed for solving Volterra integral equations of the first and second kind. The method combines support vector regression with an orthogonal kernel and Galerkin and collocation spectral methods. An optimization problem is formulated and transformed into solving a system of algebraic equations. Numerical results demonstrate the sparsity of the resulting system and the efficiency of the proposed method.
ENGINEERING WITH COMPUTERS
(2022)
Article
Engineering, Geological
Junwei Ma, Ding Xia, Haixiang Guo, Yankun Wang, Xiaoxu Niu, Zhiyang Liu, Sheng Jiang
Summary: This study proposes a hybrid approach integrating k-fold cross validation, metaheuristic support vector regression, and the nonparametric Friedman test to enhance reproducibility in landslide displacement prediction. The results demonstrate that the hybrid approach improves accuracy and reliability in ML-based prediction and recommends the use of PSO algorithm for hyperparameter tuning.
Article
Engineering, Electrical & Electronic
Xu Zhang, Nian Cai, Huan Zhang, Yun Zhang, Jianglei Di, Weisi Lin
Summary: This paper presents a novel U-shaped hybrid transformer called AFD-former for Synthesized View Quality Enhancement (SVQE). It combines the advantages of transformers and CNNs to capture global and local information collaboratively. By using the Asymmetric Flow Division Unit (AFDU), the model assigns different contributions of global-local information to the transformer and CNN branches across different layers, resulting in enhanced perceptual quality of synthesized views.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Engineering, Electrical & Electronic
Guoqing Zhang, Hongwei Zhang, Weisi Lin, Arun Kumar Chandran, Xuan Jing
Summary: Unsupervised person re-identification aims to find informative features from unlabeled person datasets. This research proposes a camera contrast learning framework, which selects camera centroids as proxies for each cluster based on time contrast principle to reduce the correlation between features and cameras. It also utilizes a 3-dimensional attention module to reduce intra-ID discrepancies caused by background shifts. Experimental results show that this method outperforms existing unsupervised person re-identification approaches on popular datasets.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Geochemistry & Geophysics
Gongyang Li, Zhi Liu, Xinpeng Zhang, Weisi Lin
Summary: In this article, a lightweight network called SeaNet is proposed for salient object detection in optical remote-sensing images (ORSI-SOD) based on semantic matching and edge alignment. SeaNet uses a lightweight MobileNet-V2 for feature extraction, a dynamic semantic matching module (DSMM) for high-level features, an edge self-alignment module (ESAM) for low-level features, and a portable decoder for inference. Experimental results demonstrate that SeaNet outperforms state-of-the-art lightweight methods and achieves comparable accuracy with conventional methods.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Computer Science, Artificial Intelligence
Wei Gao, Songlin Fan, Ge Li, Weisi Lin
Summary: Compared with current RGB or RGB-D saliency detection datasets, light field saliency detection datasets suffer from defects such as insufficient data amount and diversity, incomplete data formats, and rough annotations. To address these issues, a large-scale light field dataset called PKU-LF is constructed, which includes 5,000 light fields and covers diverse indoor and outdoor scenes. PKU-LF provides inclusive representation formats of light fields and a unified platform for comparing algorithms with different input formats.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Engineering, Civil
Hui Ying Pak, Adrian Wing-Keung Law, Weisi Lin
Summary: Water quality monitoring is essential for water resource management and governance. Remote sensing with UAVs and hyperspectral sensors has shown promise as a cost-effective and efficient method. This study has developed a new method called HBMA-OMBRA for estimating TSS concentrations from hyperspectral data, which has been verified through laboratory investigations.
JOURNAL OF HYDRO-ENVIRONMENT RESEARCH
(2023)
Article
Automation & Control Systems
Gongyang Li, Zhi Liu, Dan Zeng, Weisi Lin, Haibin Ling
Summary: In this article, a novel adjacent context coordination network (ACCoNet) is proposed for salient object detection (SOD) in optical remote sensing images (RSIs). ACCoNet improves the performance of SOD by exploring the coordination of adjacent features in an encoder-decoder architecture and introduces local and adjacent branches to handle multilevel features. Additionally, a bifurcation-aggregation block (BAB) is introduced to capture contextual information by extending the capabilities of the classic decoder block.
IEEE TRANSACTIONS ON CYBERNETICS
(2023)
Article
Engineering, Electrical & Electronic
Wen Sun, Jian Jin, Weisi Lin
Summary: Deep learning models are vulnerable to adversarial examples and existing methods focus on attacking models without considering perceptual quality. This research proposes a framework based on the MND concept for generating adversarial privacy preserving images that have minimum perceptual difference while attacking deep learning models.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Materials Science, Multidisciplinary
Qianying Feng, Jixuan Wu, Hua Bai, Binbin Song, Cheng Zhang, Wei Lin, Haifeng Liu, Shaoxiang Duan
Summary: In this study, a square coreless fiber functionalized with a Ti3C2Tx MXene layer is proposed for highly sensitive refractometric measurement. The sensitivity of the refractometric sensor is improved by more than 12% compared to the pristine fiber. The Ti3C2Tx modified square coreless fiber provides a promising platform for general ultra-low concentration analytical detection.
OPTICAL MATERIALS EXPRESS
(2023)
Article
Engineering, Electrical & Electronic
Haoning Wu, Chaofeng Chen, Liang Liao, Jingwen Hou, Wenxiu Sun, Qiong Yan, Weisi Lin
Summary: Compared with existing works, temporal relationships between frames and their influences on video quality assessment (VQA) are relatively under-studied. This study proposes a Transformer-based VQA method to tackle these issues. The method extracts spatial-temporal features and handles temporal quality attention, achieving state-of-the-art performance on multiple benchmarks.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Engineering, Electrical & Electronic
Yuxin Feng, Xiaozhe Meng, Fan Zhou, Weisi Lin, Zhuo Su
Summary: This paper proposes a sliding self-attention wavelet network for image haze removal in complex natural haze scenes. The method uses a sliding self-attention module to identify haze regions and uses discrete wavelet transform and inverse transform to construct a hierarchical encoder-decoder structure for gradually recovering sharp edges and precise texture details. Experimental results demonstrate that the proposed algorithm achieves favorable dehazing performance on relevant benchmark datasets.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Computer Science, Information Systems
Jian Xiong, Hao Gao, Miaohui Wang, Hongliang Li, King Ngi Ngan, Weisi Lin
Summary: This paper proposes an efficient geometry surface coding (EGSC) method for video point cloud compression, which improves the compression of geometry information by establishing an error projection model and using an EP-based rate-distortion optimization method. It also introduces an occupancy-map driven scheme for merge mode prediction to enhance prediction accuracy.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Artificial Intelligence
Chenlei Lv, Weisi Lin, Baoquan Zhao
Summary: In this paper, a new registration method called KSS-ICP is proposed for rigid registration in Kendall shape space (KSS) with Iterative Closest Point (ICP). The KSS is a quotient space that removes influences of translations, scales, and rotations for shape feature-based analysis. The KSS-ICP achieves accurate registration from point clouds and outperforms the state-of-the-art.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Computer Science, Information Systems
Jian Jin, Xingxing Zhang, Lili Meng, Weisi Lin, Jie Liang, Huaxiang Zhang, Yao Zhao
Summary: In this paper, an auto-weighted layer representation based view synthesis distortion estimation model is proposed, which calculates sub-synthesis distortion and learns a nonlinear mapping function to obtain the associated weights. It can efficiently and accurately estimate the view synthesis distortion.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Information Systems
Jingwen Hou, Weisi Lin, Guanghui Yue, Weide Liu, Baoquan Zhao
Summary: Personalized image aesthetics assessment aims to estimate aesthetic experiences based on individual preferences. This research proposes a method that directly estimates personalized aesthetic experiences from the interaction between image contents and user preferences, without the need for prior knowledge on generic aesthetics assessment. Extensive experiments show that the proposed method outperforms previous personalized methods and generic methods in terms of both personalized and generic aesthetics assessment.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Geochemistry & Geophysics
Huiwen Wang, Liang Liao, Jing Xiao, Weisi Lin, Mi Wang
Summary: This article proposes an enhanced remote-sensing image compression approach that utilizes uplink assistance to improve compression efficiency. By leveraging historical images from ground stations as reference images for on-orbit compression, spatiotemporal redundancy in remote-sensing images can be effectively eliminated. The proposed dual-end referencing downsampling-based coding framework effectively mitigates fake texture generation and achieves significant bitrate savings compared to standard compression baselines.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)