Theoretical Justification of a Neural Network Approach to Combinatorial Optimization
Vladislav Haralampiev · 2020
Recently a new competition-based neural network approach for solving combinatorial optimization problems was proposed. The approach demonstrates excellent performance for a number of facility location problems, including the p-MiniSum, p-Defense-Sum and p-Hub problems. Here we provide a theoretical justification of the competition-based neural network approach. It is shown that the approach asymptotically converges to an optimal solution of the problem being solved. Finite-time approximation is also discussed.