Linear Hopfield networks, inverse kinematics and constrained optimization

Methods for the design of different types of linear Hopfield networks are presented. The resulting neural networks are guaranteed to converge to their stable equilibrium, i.e. to solutions of the linear equations implicitly represented by the network. The construction of a step size is introduced, which allows convergence of the dynamic process at or near maximum rate. This work is a continuation the authors´ previous work (1994), and as an application example a neural network solution to the inverse kinematics problem is described