A primal-dual linear programming solver with linear order complexity

Recurrent artificial neural network (ANN) models are presented for solving primal-dual linear programming problems. The theoretical background is introduced based on the nonlinear analysis of an ANN. A general procedure to synthesize an ANN for optimization problems is discussed. A method to reduce the circuit complexity of the proposed ANN from the order of O(mn) to O(m+n) is developed. Simulation results are presented through an example of up to 20 variables.<
>