Article
Mathematics
Jianli Shao, Xin Liu, Wenqing He
Summary: The article introduces the use of data-adaptive SVM for instance classification in multi-class classification problems and proposes a multi-class data-dependent kernel function to enhance classification accuracy. Through simulation studies and real dataset, the excellent performance of the method is demonstrated, especially in detecting rare class instances.
Article
Computer Science, Information Systems
Jinseong Park, Yujin Choi, Junyoung Byun, Jaewook Lee, Saerom Park
Summary: In this paper, a multi-class classification method using kernel supports and a dynamical system under differential privacy is proposed. For small datasets, kernel methods, such as kernel support vector machines (SVMs), show good generalization performance with high-dimensional feature mapping. However, kernel SVMs have a fundamental weakness in achieving differential privacy because they construct decision functions based on a subset of the training data called support vectors. To address these limitations, a two-phase classification algorithm based on support vector data description (SVDD) is developed. It generates a differentially private SVDD (DP-SVDD) by perturbing the sphere center in a high-dimensional feature space and partitions the input space using a dynamical system for classification.
INFORMATION SCIENCES
(2023)
Article
Computer Science, Artificial Intelligence
Yinan Guo, Zirui Zhang, Fengzhen Tang
Summary: Feature selection is important in machine learning to reduce complexity and simplify interpretation. A novel non-linear method proposed in this paper uses kernelized multi-class support vector machines and fast recursive feature elimination to select features that work well for all classes, resulting in lower computational time complexity.
PATTERN RECOGNITION
(2021)
Article
Computer Science, Information Systems
Jie Sun, Hamido Fujita, Yujiao Zheng, Wenguo Ai
Summary: This paper focuses on multiclass financial distress prediction using SVM and decomposition fusion methods, showing that OVO-SVM outperforms OVR-SVM and ECOC-SVM in overall performance and is preferred. Data preprocessing mechanisms can greatly enhance the model performance, while OVO-SVM is more competitive for predicting financial pseudosoundness and moderate financial distress compared to human expertise.
INFORMATION SCIENCES
(2021)
Article
Computer Science, Artificial Intelligence
Wenwen Qiang, Hongjie Zhang, Jingxing Zhang, Ling Jing
Summary: The paper introduces a novel twin support vector machine, TSVM-M-3, for multi-class classification and a new RKT for large-scale classification. TSVM-M-3 considers the first and second-order moments of positive points loss and introduces an adjusting factor when constructing decision hyperplanes; RKT uses a density-dependent data selection method to reduce modeling error.
APPLIED SOFT COMPUTING
(2022)
Article
Computer Science, Artificial Intelligence
Ran An, Yitian Xu, Xuhua Liu
Summary: TSVM is suitable for STL problems, while MTL explores shared information between multiple tasks for better classification. The proposed rough MT-v-TSVM assigns different penalties to misclassified samples based on their positions, combining the advantages of rough v-TSVM and preserving the individuality of tasks.
APPLIED SOFT COMPUTING
(2021)
Article
Quantum Science & Technology
Shivani Mahashakti Pillay, Ilya Sinayskiy, Edgar Jembere, Francesco Petruccione
Summary: This article introduces a novel multi-class SWAP-Test quantum classifier that avoids the cost of developing multiple models and has invariant quantum bit requirements, measurement strategies, and circuit topologies. It is also robust under noise conditions.
ADVANCED QUANTUM TECHNOLOGIES
(2023)
Article
Computer Science, Artificial Intelligence
Liuyuan Chen, Kanglei Zhou, Junchang Jing, Haiju Fan, Juntao Li
Summary: This work proposes a fast regularization parameter tuning algorithm for the twin multi-class support vector machine. By adopting a novel sample data set partition strategy and utilizing linear equations and block matrix theory, the regularization parameters are continuously updated, and the relationship between the Lagrangian multipliers and the regularization parameters is proven. Finally, different events are defined to seek for the starting event for the next iteration, and the effectiveness of the proposed method is validated through experiments.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Zheming Gao, Shu-Cherng Fang, Xuerui Gao, Jian Luo, Negash Medhin
Summary: This paper proposes a kernel-free least squares twin support vector machine model for multi-class classification, which utilizes a special fourth order polynomial surface and one-versus-all classification strategy, with l(2) regularization to accommodate various levels of nonlinearity in datasets. Theoretical analysis and computational results demonstrate the superior performance of the proposed model, particularly for imbalanced datasets.
KNOWLEDGE-BASED SYSTEMS
(2021)
Article
Automation & Control Systems
Wei Guo, Zhe Wang, Sisi Hong, Dongdong Li, Hai Yang, Wen Du
Summary: This paper introduces a novel SVDD method - MKLSVDD, which effectively utilizes boundary information and multiple kernel learning characteristics to make use of boundary samples during the training process and designs the optimal kernel combination. Experimental results demonstrate that MKLSVDD outperforms other existing methods on multiple datasets.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2021)
Article
Computer Science, Artificial Intelligence
Liming Liu, Maoxiang Chu, Rongfen Gong, Li Zhang
Summary: The improved nonparallel support vector machine (INPSVM) proposed in this article inherits the advantages of nonparallel support vector machine (NPSVM) while also offering incomparable benefits over twin support vector machine (TSVM). INPSVM effectively eliminates noise effects and achieves higher classification accuracy for both linear and nonlinear datasets compared to other algorithms. Experimental results demonstrate the superior efficiency, accuracy, and robustness of INPSVM.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Yawen Chen, Zeyi Wen, Bingsheng He, Jian Chen
Summary: Choosing a suitable decomposition method for multi-class classification is crucial for balancing efficiency and predictive accuracy. In this paper, we present D-Chooser, an automatic method selection approach that accurately chooses the best decomposition method. D-Chooser incorporates a difficulty index that measures various aspects of the multi-class problem, and achieves an accuracy of 80.56% in selecting the best method.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Genyun Sun, Xueqian Rong, Aizhu Zhang, Hui Huang, Jun Rong, Xuming Zhang
Summary: This paper proposes a SVM classifier based on multi-scale Mahalanobis kernel, which improves the classification accuracy by optimizing parameters and enhancing global cognitive learning ability. Experimental results show that this method performs better in classifying high-resolution remote sensing images.
COGNITIVE COMPUTATION
(2021)
Article
Computer Science, Artificial Intelligence
Hong-Jie Xing, Zi-Chuan He
Summary: This study proposes a novel adaptive loss function based LS-OCSVM method to enhance its anti-outlier performance, and it demonstrates better performance on synthetic and benchmark data sets compared to nine related methods.
PATTERN RECOGNITION LETTERS
(2022)
Article
Computer Science, Artificial Intelligence
Xin Yan, Hongmiao Zhu
Summary: This paper proposes a novel support vector machine model with feature mapping and kernel trick to handle datasets with different distributions. The model improves robustness by pre-selecting training points, and converts the problem into a convex quadratic programming problem solved efficiently by the sequential minimal optimization algorithm. Numerical tests demonstrate the superior performance of the proposed method compared to other classification methods.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Biotechnology & Applied Microbiology
Hua Yang, Linmei Wang, Yingbin Zheng, Guiming Hu, Hongyan Ma, Liyun Shen
Summary: In this study, researchers discovered that ZNF267 is significantly upregulated in DLBCL and its knockdown can weaken the aggressiveness of DLBCL, reduce tumor growth, and decrease epithelial-mesenchymal transition and cancer stem cell properties.
Article
Chemistry, Physical
Jiapei Wang, Min Liu, Junjie Li, Chuanfu Wang, Xinbao Zhang, Yingbin Zheng, Xiujie Li, Longya Xu, Xinwen Guo, Chunshan Song, Xiangxue Zhu
Summary: In this study, the phase evolution of Fe-based catalysts in CO2-assisted dehydrogenation was investigated. It was found that Fe3C is the active phase for propane dehydrogenation and shows high activity for reforming and cracking when CO2 is introduced. Higher CO2/i-C4H10 feed ratio leads to a longer persistence of iron species in a higher valence state. Fe3O4 phase remains stable with a CO2/i-C4H10 feed ratio of 1/2 or higher, resulting in a relatively stable isobutane conversion.
Article
Computer Science, Artificial Intelligence
Zhichao Fu, Yingbin Zheng, Tianlong Ma, Hao Ye, Jing Yang, Liang He
Summary: In this paper, a two-phase edge-aware deep network is proposed to improve deep image deblurring, and promising results are achieved.
Article
Immunology
Xinyue Tu, Jing Yang, Yingbin Zheng, Chen Liang, Qiang Tao, Xiang Tang, Zonghao Liu, Lingmin Jiang, Zhaoqian He, Feihu Xie, Yun Zheng
Summary: This study compared the efficacy and safety of immunotherapy plus regorafenib with regorafenib alone in patients with pretreated hepatocellular carcinoma. The results showed that immunotherapy plus regorafenib significantly improved clinical outcomes and had a manageable safety profile compared with regorafenib monotherapy in advanced HCC after front-line therapy failure.
INTERNATIONAL IMMUNOPHARMACOLOGY
(2022)
Article
Engineering, Chemical
Wen Liu, Junjie Li, Qiang Yu, Yanan Wang, Weifeng Chu, Yingbin Zheng, Zhiqiang Yang, Xuebin Liu, Xiujie Li, Xiangxue Zhu
Summary: This study developed a direct synthesis route for submicron (500 nm) spherical ZSM-48 zeolite with the assistance of F- ions at low crystallization temperature. Through detailed investigation, a reliable formation mechanism of spherical ZSM-48 zeolite was proposed. The submicron spherical ZSM-48 zeolite showed significantly improved stability and higher selectivity for p-xylene in m-xylene isomerization compared to traditional zeolite.
INDUSTRIAL & ENGINEERING CHEMISTRY RESEARCH
(2023)
Article
Neurosciences
Kai Gong, Qian Dai, Jiacheng Wang, Yingbin Zheng, Tao Shi, Jiaxing Yu, Jiangwang Chen, Shaohui Huang, Zhanxiang Wang
Summary: With the development of deep learning, Computer-Aided Diagnosis (CAD) tasks in emergency medicine using Non-Contrast head Computed Tomography (NCCT) for IntraCerebral Hematoma (ICH) have gained popularity. However, challenges such as time-consuming manual evaluation, high cost patient-level predictions, and the need for high accuracy and interpretability remain. This paper proposes a multi-task framework to address these challenges, consisting of a weight-shared module for feature extraction and two heads for regression and classification tasks. Experimental results demonstrate that the multi-task framework outperforms the single-task framework in terms of performance and interpretability using Grad-CAM.
FRONTIERS IN NEUROSCIENCE
(2023)
Article
Computer Science, Artificial Intelligence
Xiangcheng Du, Zhao Zhou, Yingbin Zheng, Xingjiao Wu, Tianlong Ma, Cheng Jin
Summary: This paper presents a method based on self-supervised learning, which uses unlabeled real-world scene text images for feature representation. A novel pretext task is designed to ensure consistency among text stroke masks of image variants. The Progressive Erasing Network is employed to remove residual texts progressively, leveraging intermediate generated results for higher quality results. Experimental results show that this method significantly improves the generalization of the text erasing task and achieves state-of-the-art performance on public benchmarks.
COMPUTER VISION AND IMAGE UNDERSTANDING
(2023)
Article
Health Care Sciences & Services
Xiaobo Zhang, Ying Gu, Jie Yin, Yuejie Zhang, Cheng Jin, Weibing Wang, Albert Martin Li, Yingwen Wang, Ling Su, Hong Xu, Xiaoling Ge, Chengjie Ye, Liangfeng Tang, Bing Shen, Jinwu Fang, Daoyang Wang, Rui Feng
Summary: This study aims to develop a comprehensive scale for measuring the knowledge, attitude, and practice of ethics implementation among medical AI researchers, and to evaluate its measurement properties. The results demonstrate that the scale has good reliability and structural validity, making it an effective instrument for assessing ethics implementation in medical AI research.
JMIR FORMATIVE RESEARCH
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Hao Zhang, Xin Chen, Heming Jing, Yingbin Zheng, Yuan Wu, Cheng Jin
Summary: Visual place recognition is improved with an Efficient Transformer for Re-ranking (ETR), which uses both global and local descriptors to re-rank the top candidates in a single shot. The ETR leverages self-attention and cross-attention to capture relationships within a single image and explore the similarity between image pairs. Experimental results demonstrate that ETR outperforms existing methods in terms of computational efficiency and retrieval performance.
2023 IEEE/CVF WINTER CONFERENCE ON APPLICATIONS OF COMPUTER VISION (WACV)
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Xiangcheng Du, Zhao Zhou, Yingbin Zheng, Tianlong Ma, Xingjiao Wu, Cheng Jin
Summary: A new end-to-end network is proposed in this paper, which focuses on modeling text stroke masks to compute more accurate locations for erased images. The network consists of two stages, a basic network and a refinement network. The basic network predicts text stroke masks and initial erasing results simultaneously. The refinement network generates natural erased results using the masks as supervision. Experimental results demonstrate the effectiveness of our framework in producing high-quality erasing results.
2023 IEEE/CVF WINTER CONFERENCE ON APPLICATIONS OF COMPUTER VISION (WACV)
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Jingwen Ye, Yining Mao, Jie Song, Xinchao Wang, Cheng Jin, Mingli Song
Summary: This paper proposes a framework called Safe Distillation Box (SDB) for protecting the ownership of a pre-trained model. SDB wraps the model to prevent unauthorized knowledge distillation, while enhancing knowledge for authorized users. Experiments demonstrate the effectiveness of SDB.
THIRTY-SIXTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE / THIRTY-FOURTH CONFERENCE ON INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE / THE TWELVETH SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Kaiyi Zhang, Ximing Yang, Yuan Wu, Cheng Jin
Summary: In this work, we propose a novel attention-based method, AXform, for transforming latent features to point clouds. By considering parameter sharing and data flow, our method generates point clouds with fewer outliers, fewer network parameters, and faster convergence speed. Furthermore, it also performs well in generating non-smooth surfaces and enabling unsupervised semantic segmentation.
THIRTY-SIXTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE / THIRTY-FOURTH CONFERENCE ON INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE / THE TWELVETH SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Hao Ren, Haoran Ren, Wu Ran, Hong Lu, Cheng Jin
Summary: This paper proposes a method based on a multi-head cross-modal attention mechanism, which enhances RGB and Flow features by considering the cross-correlation between different modal features. Experimental results demonstrate that this method can significantly improve the performance of existing methods and achieve state-of-the-art results on two datasets.
PRICAI 2022: TRENDS IN ARTIFICIAL INTELLIGENCE, PT III
(2022)
Proceedings Paper
Computer Science, Artificial Intelligence
Yuan Wu, Yanlu Cai, Rui Feng, Cheng Jin
Summary: Spatial and temporal information, as well as multi-view mutual promotion, are crucial for addressing depth uncertainty in 2D-to-3D human pose estimation. This study presents a novel multi-view multiframe method called (STPE)-P-2, which combines a Single-view Spatial and Temporal Transformer (S-ST2) and a Multiple Cross-view Transformer-based Transmission (M-CT2) module. Experimental results on three mainstream datasets demonstrate that (STPE)-P-2 achieves state-of-the-art performance among existing 2D-to-3D methods.
ARTIFICIAL NEURAL NETWORKS AND MACHINE LEARNING - ICANN 2022, PT II
(2022)