Toward Deep Neural Networks: WASD Neuronet Models, Algorithms, and Applications introduces the outlook and extension toward deep neural networks, with a focus on the weights-and-structure determination (WASD) algorithm. Based on the authors’ 20 years of research experience on neuronets, the book explores the models, algorithms, and applications of the WASD neuronet, and allows reader to extend the techniques in the book to solve scientific and engineering problems. The book will be of interest to engineers, senior undergraduates, postgraduates, and researchers in the fields of neuronets, computer mathematics, computer science, artificial intelligence, numerical algorithms, optimization, simulation and modeling, deep learning, and data mining. Features Focuses on neuronet models, algorithms, and applications Designs, constructs, develops, analyzes, simulates and compares various WASD neuronet models, such as single-input WASD neuronet models, two-input WASD neuronet models, three-input WASD neuronet models, and general multi-input WASD neuronet models for function data approximations Includes real-world applications, such as population prediction Provides complete mathematical foundations, such as Weierstrass approximation, Bernstein polynomial approximation, Taylor polynomial approximation, and multivariate function approximation, exploring the close integration of mathematics (i.e., function approximation theories) and computers (e.g., computer algorithms) Utilizes the authors' 20 years of research on neuronets

This book introduces readers to using the simple but effective Zhang-gradient (ZG) method to solve tracking-control problems concerning various nonlinear systems, while also highlighting the applications of the ZG method to tracking control for practical systems, e.g. an inverted-pendulum-on-a-cart (IPC) system and a two-wheeled mobile robot (showing its potential applications). In addition to detailed theoretical analyses of ZG controllers, the book presents a wealth of computer simulations to demonstrate the feasibility and efficacy of the controllers discussed (as well as the method itself). More importantly, the superiority of ZG controllers in overcoming the division-by-zero (DBZ) problem is also illustrated. Given its scope and format, the book is well suited for undergraduate and graduate students, as well as academic and industrial researchers in the fields of neural dynamics/neural networks, nonlinear control, computer mathematics, time-varying problem solving, modeling and simulation, analog hardware, and robotics.

Repetitive Motion Planning and Control of Redundant Robot Manipulators presents four typical motion planning schemes based on optimization techniques, including the fundamental RMP scheme and its extensions. These schemes are unified as quadratic programs (QPs), which are solved by neural networks or numerical algorithms. The RMP schemes are demonstrated effectively by the simulation results based on various robotic models; the experiments applying the fundamental RMP scheme to a physical robot manipulator are also presented. As the schemes and the corresponding solvers presented in the book have solved the non-repetitive motion problems existing in redundant robot manipulators, it is of particular use in applying theoretical research based on the quadratic program for redundant robot manipulators in industrial situations. This book will be a valuable reference work for engineers, researchers, advanced undergraduate and graduate students in robotics fields. Yunong Zhang is a professor at The School of Information Science and Technology, Sun Yat-sen University, Guangzhou, China; Zhijun Zhang is a research fellow working at the same institute.

This book focuses on solving different types of time-varying problems. It presents various Zhang dynamics (ZD) models by defining various Zhang functions (ZFs) in real and complex domains. It then provides theoretical analyses of such ZD models and illustrates their results. It also uses simulations to substantiate their efficacy and show the feasibility of the presented ZD approach (i.e., different ZFs leading to different ZD models), which is further applied to the repetitive motion planning (RMP) of redundant robots, showing its application potential.

"Repetitive Motion Planning and Control of Redundant Robot Manipulators" presents four typical motion planning schemes based on optimization techniques, including the fundamental RMP scheme and its extensions. These schemes are unified as quadratic programs (QPs), which are solved by neural networks or numerical algorithms. The RMP schemes are demonstrated effectively by the simulation results based on various robotic models; the experiments applying the fundamental RMP scheme to a physical robot manipulator are also presented. As the schemes and the corresponding solvers presented in the book have solved the non-repetitive motion problems existing in redundant robot manipulators, it is of particular use in applying theoretical research based on the quadratic program for redundant robot manipulators in industrial situations. This book will be a valuable reference work for engineers, researchers, advanced undergraduate and graduate students in robotics fields.

Yunong Zhang is a professor at The School of Information Science and Technology, Sun Yat-sen University, Guangzhou, China; Zhijun Zhang is a research fellow working at the same institute.

"

Toward Deep Neural Networks: WASD Neuronet Models, Algorithms, and Applications introduces the outlook and extension toward deep neural networks, with a focus on the weights-and-structure determination (WASD) algorithm. Based on the authors' 20 years of research experience on neuronets, the book explores the models, algorithms, and applications of the WASD neuronet, and allows reader to extend the techniques in the book to solve scientific and engineering problems. The book will be of interest to engineers, senior undergraduates, postgraduates, and researchers in the fields of neuronets, computer mathematics, computer science, artificial intelligence, numerical algorithms, optimization, simulation and modeling, deep learning, and data mining. Features Focuses on neuronet models, algorithms, and applications Designs, constructs, develops, analyzes, simulates and compares various WASD neuronet models, such as single-input WASD neuronet models, two-input WASD neuronet models, three-input WASD neuronet models, and general multi-input WASD neuronet models for function data approximations Includes real-world applications, such as population prediction Provides complete mathematical foundations, such as Weierstrass approximation, Bernstein polynomial approximation, Taylor polynomial approximation, and multivariate function approximation, exploring the close integration of mathematics (i.e., function approximation theories) and computers (e.g., computer algorithms) Utilizes the authors' 20 years of research on neuronets

Neural networks and neural dynamics are powerful approaches for the online solution of mathematical problems arising in many areas of science, engineering, and business. Compared with conventional gradient neural networks that only deal with static problems of constant coefficient matrices and vectors, the authors' new method called zeroing dynamics solves time-varying problems. Zeroing Dynamics, Gradient Dynamics, and Newton Iterations is the first book that shows how to accurately and efficiently solve time-varying problems in real-time or online using continuous- or discrete-time zeroing dynamics. The book brings together research in the developing fields of neural networks, neural dynamics, computer mathematics, numerical algorithms, time-varying computation and optimization, simulation and modeling, analog and digital hardware, and fractals. The authors provide a comprehensive treatment of the theory of both static and dynamic neural networks. Readers will discover how novel theoretical results have been successfully applied to many practical problems. The authors develop, analyze, model, simulate, and compare zeroing dynamics models for the online solution of numerous time-varying problems, such as root finding, nonlinear equation solving, matrix inversion, matrix square root finding, quadratic optimization, and inequality solving.

The volume LNCS 9377 constitutes the refereed proceedings of the 12th International Symposium on Neural Networks, ISNN 2015, held in Jeju, South Korea in October 2015. The 55 revised full papers presented were carefully reviewed and selected from 97 submissions. These papers cover many topics of neural network-related research including intelligent control, neurodynamic analysis, memristive neurodynamics, computer vision, signal processing, machine learning, and optimization.

The real-time solution to a mathematical problem arises in numerous fields of science, engineering, and business. It is usually an essential part of many solutions, e.g., matrix/vector computation, optimisation, control theory, kinematics, signal processing, and pattern recognition. In recent years, due to the in-depth research on neural networks, numerous recurrent neural networks (RNN) based on the gradient-based method have been developed and investigated. Particularly, some simple neural networks were proposed to solve linear programming problems in real time and implemented on analogue circuits. In this book, ZNN, ZD or ZND theory formalises these problems and solutions in the time-varying context and provides compact models that could solve those dynamic problems.

Because of massively parallel distributed nature and very fast convergence rates, recurrent neural networks (RNN) are widely applied to solving many problems in optimization, control and robotic systems, etc. Hence, this book investigates the following RNN models which solve some practical problems, together with their corresponding analysis on stability and convergence. A type of multilayer pole-assignment neural networks is applied to online synthesizing and tuning feedback control systems. Then, a novel RNN model is established by absorbing the first-order time-derivative information to solve the Sylvester equation with time-varying coefficient matrices. A dual neural network is developed to handle quadratic programs subject to linear constraints. The Lagrangian neural network and primal-dual neural network are also reviewed for comparison purposes. The neural networks are then exploited for real-time motion planning of redundant manipulators. The publication is primarily intended for researchers and postgraduates studying in RNN, control and robotics.