A fast algorithm for solving large scale nonlinear optimization problems using RNN

This paper presents a discrete-time recurrent neural network (RNN) model for solving nonlinear differentiable constrained optimization problems, which contain the special case of convex optimizations over constrained sets and variational inequality problem. The qualitative analysis results about the regularity and completeness of the proposed network have been obtained. It is shown that all trajectories starting from any initial point in Rfrⁿ converge to the equilibrium set of the recurrent system. This RNN model shows its great simplicity in contrast to other existing neural network solvers. Simulations for a class of large scale linear complementarity problems illustrate the fast convergence and features of the proposed RNN model.