Feasible directions linear programming by neural networks
Valmir Carneiro Barbosa, Luı́s Alfredo Vidal de Carvalho · 1990
The authors describe how a neural network can be built for the exact solution of linear programming problems by a feasible directions approach. The proposed network, when started at any interior point, continuously tracks a path of equally interior points converging to an optimal solution. The number of neurons in the network grows linearly with the problem size, and neurons are relatively sparsely connected. Initial simulation results indicate that convergence is very fast, and consequently the network can be of relevance to many application areas