Senbo Fu - Academia.edu (original) (raw)
Uploads
Papers by Senbo Fu
By following Zhang et al.'s design method, a special class of recurrent neural network termed Zha... more By following Zhang et al.'s design method, a special class of recurrent neural network termed Zhang neural network (ZNN) has been proposed for online solution of time-varying linear inequalities. For the purpose of digital-hardware implementation, the resultant ZNN model is discretized by employing Euler difference rule in this paper. Thus, three discrete-time ZNN models and numerical algorithms (i.e., discrete-time ZNN algorithms, in short) are proposed and investigated for online solution of time-varying linear matrix-vector inequalities. In addition, a criterion is proposed to measure the rapidity and accuracy of the proposed discrete-time ZNN algorithms. Numerical-study results further verify and demonstrate the efficacy of the proposed discrete-time ZNN algorithms for online solution of time-varying linear matrix-vector inequalities.
Since March 2001, a special class of recurrent neural network termed Zhang neural network (ZNN) h... more Since March 2001, a special class of recurrent neural network termed Zhang neural network (ZNN) has been proposed by Zhang et al for time-varying matrix inversion. For the purpose of possible hardware implementation, the resultant ZNN model is discretized by employing Euler forward-difference rule. In this paper, three discrete-time ZNN models using nonlinear activation functions (e.g., power-sigmoid activation functions) are presented and investigated for time-varying matrix inversion. In addition, a criterion is proposed to measure the rapidity and accuracy of the presented discrete-time ZNN models for time-varying matrix inversion. Numerical results further demonstrate the efficacy of the presented discrete-time ZNN models for time-varying matrix inversion.
In this paper, a numerical method (termed, E47 algorithm) based on linear variational inequalitie... more In this paper, a numerical method (termed, E47 algorithm) based on linear variational inequalities (LVI) is presented and investigated to solve quadratic programming (QP) problems which are simultaneously subject to linear equality, inequality and bound constraints. Note that such constrained QP problems can be equivalent to linear variational inequalities and then to piecewise-linear projection equations (PLPE). The E47 algorithm is then adapted to solving the resultant PLPE, and thus the optimal numerical solutions to the QP problems are obtained. In addition, the global linear convergence of such an E47 algorithm is proved. The numerical comparison results between such an E47 algorithm and the active set algorithm are further provided. The efficacy and superiority of the presented E47 algorithm for QP solving are substantiated.
By following Zhang et al.'s design method, a special class of recurrent neural network termed Zha... more By following Zhang et al.'s design method, a special class of recurrent neural network termed Zhang neural network (ZNN) has been proposed for online solution of time-varying linear inequalities. For the purpose of digital-hardware implementation, the resultant ZNN model is discretized by employing Euler difference rule in this paper. Thus, three discrete-time ZNN models and numerical algorithms (i.e., discrete-time ZNN algorithms, in short) are proposed and investigated for online solution of time-varying linear matrix-vector inequalities. In addition, a criterion is proposed to measure the rapidity and accuracy of the proposed discrete-time ZNN algorithms. Numerical-study results further verify and demonstrate the efficacy of the proposed discrete-time ZNN algorithms for online solution of time-varying linear matrix-vector inequalities.
Since March 2001, a special class of recurrent neural network termed Zhang neural network (ZNN) h... more Since March 2001, a special class of recurrent neural network termed Zhang neural network (ZNN) has been proposed by Zhang et al for time-varying matrix inversion. For the purpose of possible hardware implementation, the resultant ZNN model is discretized by employing Euler forward-difference rule. In this paper, three discrete-time ZNN models using nonlinear activation functions (e.g., power-sigmoid activation functions) are presented and investigated for time-varying matrix inversion. In addition, a criterion is proposed to measure the rapidity and accuracy of the presented discrete-time ZNN models for time-varying matrix inversion. Numerical results further demonstrate the efficacy of the presented discrete-time ZNN models for time-varying matrix inversion.
In this paper, a numerical method (termed, E47 algorithm) based on linear variational inequalitie... more In this paper, a numerical method (termed, E47 algorithm) based on linear variational inequalities (LVI) is presented and investigated to solve quadratic programming (QP) problems which are simultaneously subject to linear equality, inequality and bound constraints. Note that such constrained QP problems can be equivalent to linear variational inequalities and then to piecewise-linear projection equations (PLPE). The E47 algorithm is then adapted to solving the resultant PLPE, and thus the optimal numerical solutions to the QP problems are obtained. In addition, the global linear convergence of such an E47 algorithm is proved. The numerical comparison results between such an E47 algorithm and the active set algorithm are further provided. The efficacy and superiority of the presented E47 algorithm for QP solving are substantiated.