Neural network architecture for linear programming
Thomas P. Caudell, Karel Zikan · 2003
A neural network architecture, called LP-Net, is introduced that rapidly solves general linear programming problems. Mathematically, the approach is based on the logarithmic barrier function approach to linear programming. The neural network simulates the barrier method's first-order dynamic system. The authors briefly outline the logarithmic barrier technique, present the neural network architecture, and give the set of differential equations that describes the network dynamics. The convergence properties of this neural network makes it ideal for analog hardware implementation.>