A neutral net linear programming solver

A linear programming solver which can find the optimal solution for primal-dual linear programming (PDLP) problems is presented. The neural net solver is based on an extension model of the Kosko´s bidirectional associative memory (BAM). The proposed neural net solver uses the same number of neurons and connections as does the Hopfield model or Chua´s canonical nonlinear programming model while it resolves the problems of (1) getting infeasible solutions and (2) employing a large penalty term in the objective function. Simulation results based on RC-active circuit implementation of the neural net solver are presented. The results indicate that when the solver is implemented in hardware with full parallelism, it is capable of solving linear programming problems very quickly and this computation speed seems to be insensitive to the dimension of the underlying problem