Article
Computer Science, Artificial Intelligence
Yang Si, Difeng Wang, Yao Chou, Dongyang Fu
Summary: This study proposes a finite-time convergent non-convex zeroing neural network (FT-NCZNN) model, which addresses the limitations of existing zeroing neural network (ZNN) models related to convex and unsaturation constraints by introducing a projection method for constructing nonconvex activation functions. Rigorous theoretical analyses verify the convergence of the model and provide an upper boundary on convergence time. Simulations demonstrate the superior performance of the FT-NCZNN model under different noise conditions. An engineering application on motion generation further illustrates the validity and feasibility of the FT-NCZNN model.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Lin Xiao, Wenqian Huang, Lei Jia, Xiaopeng Li
Summary: This paper proposes two discrete nonlinear and noise-tolerant ZNN models to solve the time-varying augmented complex Sylvester equation. The theoretical analysis and simulative experiments demonstrate the superior performance of these models.
Article
Computer Science, Artificial Intelligence
Ratikanta Behera, Dimitris Gerontitis, Predrag Stanimirovic, Vasilios Katsikis, Yang Shi, Xinwei Cao
Summary: Defining efficient families of RNN models for solving time-varying nonlinear equations is an interesting research topic in applied mathematics. The use of efficient activation functions is a key element in designing RNN. This study proposes a new family of activation functions and corresponding ZNN models, called VPIZNN, which demonstrate accelerated finite-time convergence for time-varying nonlinear equations. Theoretical results and numerical experiments support the superiority of the proposed VPIZNN models in solving the Van der Pol equation and finding the root (m)va(t).
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Mathematics, Applied
Huamin Zhang, Hongcai Yin
Summary: This research discusses the time-varying solution of a generalized linear matrix equation with the transpose of an unknown matrix. A computation model is constructed using the zeroing neural network method, and asymptotic convergence proof is provided. The study also explores the activation function with predefined-time convergence property and noise suppression strategy.
Article
Mathematics
Houssem Jerbi, Hadeel Alharbi, Mohamed Omri, Lotfi Ladhar, Theodore E. Simos, Spyridon D. Mourtas, Vasilios N. Katsikis
Summary: This article introduces the hyperpower family of iterative methods, which is widely used for approximating various matrix equation problems. It also presents the zeroing neural network (ZNN), a type of neural dynamics designed for handling time-varying problems. By analogy, a family of ZNN models that correlate with the hyperpower iterative methods, known as higher-order ZNN models (HOZNN), is defined to find real symmetric solutions of time-varying algebraic Riccati equations. Furthermore, a noise-handling HOZNN (NHOZNN) class of dynamical systems is introduced, and the traditional ZNN and HOZNN dynamic flows are compared theoretically and numerically.
Article
Automation & Control Systems
Theodore E. E. Simos, Vasilios N. N. Katsikis, Spyridon D. D. Mourtas, Predrag S. S. Stanimirovic
Summary: The problem of solving algebraic Riccati equations and certain linear matrix equations which arise from the ARE frequently occur in applied and pure mathematics, science, and engineering applications. In this article, the time-varying NARE problem is proposed and investigated by considering the nonsymmetric ARE as a general form of ARE. The effectiveness of the proposed dynamical systems is proven in ten numerical experiments, three of which include applications to LTV and nonlinear systems.
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS
(2023)
Article
Mathematics, Applied
Dimitrios Gerontitis, Ratikanta Behera, Panagiotis Tzekis, Predrag Stanimirovic
Summary: A family of varying-parameter finite-time zeroing neural networks (VPFTZNN) is proposed for solving the time-varying Sylvester equation (TVSE), with analysis of convergence speed, stability, and noise resistance, and experimental verification of theoretical results.
JOURNAL OF COMPUTATIONAL AND APPLIED MATHEMATICS
(2022)
Article
Automation & Control Systems
Zeshan Hu, Kenli Li, Lin Xiao, Yaonan Wang, Mingxing Duan, Keqin Li
Summary: This article introduces two ADTIZD models to address the TVCSE problem, showcasing higher accuracy and lower complexity. The convergence and robustness of ADTIZD models are rigorously proven and supported by numerical experiments, with comparisons to other neural network models presented. The efficacy of ADTIZD models is validated through simulation in controlling a robotic manipulator.
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS
(2022)
Article
Computer Science, Information Systems
Yihui Lei, Jiamei Luo, Tengxiao Chen, Lei Ding, Bolin Liao, Guangping Xia, Zhengqi Dai
Summary: This paper proposes a nonlinear activated integration-enhanced neural network model based on a coalescent activation function. The model accelerates convergence speed without significant efficiency loss and solves the time-varying Sylvester equation in various noise situations.
Article
Mathematics, Applied
Yunong Zhang, Xiao Liu, Yihong Ling, Min Yang, Huanchang Huang
Summary: This paper introduces a continuous ZD (CZD) model and three discrete ZD (DZD) models to solve the problem of time-varying Sylvester-transpose matrix inequality, with theoretical truncation error analyses and numerical experiments confirming the convergence, efficacy, and superiority of the DZD models.
NUMERICAL ALGORITHMS
(2021)
Article
Computer Science, Artificial Intelligence
Qing Hu, Bing Zheng
Summary: In this article, an efficient Takagi-Sugeno fuzzy zeroing neural network (TS-FZNN) with a new activation function is proposed to solve the time-varying Sylvester equation. The model's convergence and robustness are analyzed, and theoretical analysis shows that it has faster fixed convergence time compared to other ZNN models and is noise-tolerant. Numerical experiments demonstrate its efficiency and effectiveness for solving the time-varying Sylvester equation and its applicability to robot manipulator.
IEEE TRANSACTIONS ON FUZZY SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Qiuyue Zuo, Kenli Li, Lin Xiao, Keqin Li
Summary: Two robust finite-time zeroing neural network models are generated using an improved activation function, which improves convergence speed and robustness. Through comparative experiments and application to robots, the effectiveness of this method is demonstrated.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Shuqiao Wang, Long Jin, Xiujuan Du, Predrag S. Stanimirovi
Summary: The paper introduces a zeroing neurodynamics approach for solving multi-linear systems with di-tensors, proposing three specific models that converge in finite time and investigating activation functions needed for constructing these models. Theoretical analyses demonstrate the stability of the proposed approach and the convergence of the models to the theoretical solution in finite time.
Article
Computer Science, Artificial Intelligence
Dimitrios Gerontitis, Ratikanta Behera, Yang Shi, Predrag S. Stanimirovic
Summary: A robust noise-tolerant zeroing neural network (ZNN) is proposed for solving time-varying linear matrix equations (TVLME). The convergence speed of the designed neural dynamics is analyzed and compared with traditional activation functions. The proposed activation is utilized in nonlinear ZNN dynamics to solve time-varying linear matrix equations and the Stein equation. The behavior of the proposed robust noise-tolerant ZNN is investigated theoretically and experimentally. Simulation tests show the effectiveness of the suggested activation over existing activation functions.
Article
Mathematics, Applied
Zhongbo Sun, Yongbai Liu, Gang Wang, Yufeng Lian, Keping Liu, Long Jin
Summary: This paper revisits the zeroing-dynamic design formula and continuous time Z-type model for solving nonlinear time-varying equation problems (NTVEPs), and develops a modified Z-type design formula to address NTVEPs in the presence of noises. It proposes a novel class of discrete-time noise-tolerant Z-type models and general-type models, demonstrating their stability, consistency, convergence, efficiency, and robustness. The numerical results show the efficacy and superiority of the proposed models for noise-polluted NTVEPs compared with classical methods.
JOURNAL OF COMPUTATIONAL AND APPLIED MATHEMATICS
(2022)
Article
Mathematics, Applied
Yunong Zhang, Xiao Liu, Yihong Ling, Min Yang, Huanchang Huang
Summary: This paper introduces a continuous ZD (CZD) model and three discrete ZD (DZD) models to solve the problem of time-varying Sylvester-transpose matrix inequality, with theoretical truncation error analyses and numerical experiments confirming the convergence, efficacy, and superiority of the DZD models.
NUMERICAL ALGORITHMS
(2021)
Article
Engineering, Electrical & Electronic
Fuzhen Zhuang, Zhiyuan Qi, Keyu Duan, Dongbo Xi, Yongchun Zhu, Hengshu Zhu, Hui Xiong, Qing He
Summary: Transfer learning aims to improve the performance of target learners by transferring knowledge from related source domains, reducing the reliance on target-domain data. This survey aims to systematize and summarize existing research studies in order to help readers understand the current status and ideas in the area of transfer learning.
PROCEEDINGS OF THE IEEE
(2021)
Article
Computer Science, Artificial Intelligence
Min Yang, Yunong Zhang, Haifeng Hu
Summary: This paper investigates posture control of a two-manipulator system, proposing two posture control schemes based on kinematics and control tasks for each manipulator. A posture coordination control scheme is then proposed using standard quadratic programming, along with the development of a projection neural network (PNN) model. Experiments verify the effectiveness of the posture coordination control scheme, PNN model, and a one-iteration DPNN model for numerical algorithm development and digital hardware implementation.
Article
Mathematics, Applied
Min Yang, Yunong Zhang, Haifeng Hu
Summary: The paper investigates the relationship between the number of time instants and the precision of ZeaD formulas, finding that different number of instants lead to different precisions including linear, quadratic, cubic, quartic, and quintic. Theoretical analysis substantiates the relationship, while the application of ZeaD formulas to solving time-varying quadratic optimization problem demonstrates their effectiveness with numerical results.
NUMERICAL ALGORITHMS
(2021)
Article
Computer Science, Artificial Intelligence
Yunong Zhang, Yihong Ling, Min Yang, Song Yang, Zhijun Zhang
Summary: This article explores the time-varying matrix pseudoinverse (TVMP) problem and the future matrix pseudoinverse (FMP) problem, proposing new discrete ZNN models for computing the pseudoinverses of various full-rank matrices, including the Zhang matrix. Numerical experiments confirm the effectiveness and superiority of the newly proposed models.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Min Yang, Yunong Zhang, Xuefeng Zhou, Haifeng Hu
Summary: The importance of accuracy and safety in redundant arm control is highlighted in this study, with a focus on considering physical constraints for safe operation. A pose control scheme incorporating physical constraints is proposed using zeroing neural network method and quadratic programming, with a projection neural network solver and one-iteration computing algorithm developed for efficient computing and real-time requirements. Experimental results support the effectiveness and superiority of the proposed scheme and algorithms in controlling redundant arm.
APPLIED SOFT COMPUTING
(2021)
Article
Automation & Control Systems
Min Yang, Yunong Zhang, Ning Tan, Haifeng Hu
Summary: In this article, a concise continuous ZNN (CZNN) controller and two discrete ZNN (DZNN) controllers are proposed based on zeroing neural network (ZNN), which combine the end-effector tracking task and the obstacle avoidance task to guarantee the safety and effectiveness of redundant manipulators.
IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS
(2022)
Article
Computer Science, Information Systems
Yongchun Zhu, Fuzhen Zhuang, Xiangliang Zhang, Zhiyuan Qi, Zhiping Shi, Juan Cao, Qing He
Summary: Many few-shot learning approaches have been designed under the meta-learning framework, but they often suffer from data shift in real-world applications. In this paper, we propose a metric-based meta-learning framework using knowledge graph to address the data shift problem within/between tasks, and achieve remarkable performance in evaluations.
FRONTIERS OF COMPUTER SCIENCE
(2023)
Article
Automation & Control Systems
Min Yang, Yunong Zhang, Ning Tan, Mingzhi Mao, Haifeng Hu
Summary: This study researches future different-level linear matrix systems and proposes continuous different-level linear matrix systems as well as zeroing neural network equivalency. The author obtains a 7-instant discrete-time synthesis model using high-precision discretization formulas, and validates its effectiveness in solving future different-level linear matrix systems.
IEEE TRANSACTIONS ON CYBERNETICS
(2022)
Article
Automation & Control Systems
Min Yang, Yunong Zhang, Zhijun Zhang, Haifeng Hu
Summary: This article introduces a 6-step DZNN model for repetitive motion control of redundant manipulators. The theoretical analyses and computer simulations verify its efficacy, and physical experiments on the Kinova Jaco(2) manipulator demonstrate its practicality.
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS
(2022)
Article
Automation & Control Systems
Yunong Zhang, Min Yang, Huanchang Huang, Jianrong Chen, Zhonghua Li
Summary: This article investigates time-varying matrix equation problems from a future perspective, proposing a nine-instant DTUS model for a unified solution. The theoretical analyses and numerical experiments conducted validate the effectiveness and superiority of the 9IDTUS model.
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS
(2022)
Article
Automation & Control Systems
Min Yang, Yunong Zhang, Ning Tan, Haifeng Hu
Summary: In this article, explicit linear left-and-right 5-step (ELLR5S) formulas with sixth-order precision are proposed for time discretization. The formulas are utilized to solve time-varying linear and nonlinear systems and are shown to be effective and superior through theoretical analyses and experiments.
IEEE TRANSACTIONS ON CYBERNETICS
(2023)
Article
Computer Science, Artificial Intelligence
Yunong Zhang, Zhenyu Li, Min Yang, Liangjie Ming, Jinjin Guo
Summary: Equivalency is a powerful approach that can transform a problem into a more solvable form. The study introduces Zhang neurodynamics equivalency (ZNE) as a process that can solve equations and inequations. The application of ZNE in robot motion planning and control demonstrates its effectiveness.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Proceedings Paper
Engineering, Industrial
Zanyu Tang, Zhiyuan Qi, Yunong Zhang
Summary: This paper summarizes the Zhang neural network (ZNN) models and Zhang time discretization (ZTD) formulas for solving time-dependent problems. It provides various choices of ZNN models and ZTD formulas for different time-dependent problems. The paper categorizes and presents seventeen subcategories of time-dependent problems, and compares different ZTD formulas. It also introduces the process of solving time-dependent problems using the ZNN method.
2022 IEEE 17TH CONFERENCE ON INDUSTRIAL ELECTRONICS AND APPLICATIONS (ICIEA)
(2022)
Article
Computer Science, Artificial Intelligence
Min Yang, Yunong Zhang, Haifeng Hu
Summary: This paper proposes an inverse-free continuous ZNN model for solving time-dependent linear systems, and develops a discrete model using a general linear six-step method. The proposed discrete model achieves higher precision and improvement compared to conventional discretization methods. The efficacy and applicability of the models are demonstrated through theoretical analysis and specific examples.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
