Application of projection neural network in solving convex programming problems (Q2371989)

Two types of methods are commonly used to formulate an optimization problem in terms of a neural network. The first approach converts the constrained optimization problem into an associated unconstrained optimization problem and a neural network is designed to solve the unconstrained problem with gradient methods. The second approach constructs a set of differential equations such that their equilibrium points correspond to the desired solution and an appropriate Lyapunov function is found such that all trajectories of the system converge to some equilibrium points. The authors combine the two approaches by introducing neural network models for solving the nonlinear projection formulation. The stability condition and convergence are analyzed, and six numerical examples are presented.