New approach for solving the travelling salesman problem using self-organizing learning

In applying Kohonen's self-organizing model to solve travelling salesman problem (TSP), it is observed that the quality of the solution depends on the number of neurons being used, which is around two to three times of the number of cities, and is highly problem dependent. Instead of doing extensive experiments to determine the optima number of neurons, we propose a new winner selection criterion which generalizes the conventional ones. With this criterion, a ring of N neurons for solving an N-city TSP gives better solution as compared to those given by rings with N, 2N, 3N and 4N neurons using the conventional selection criterion, and yet takes only 16% more processing time than that of the conventional approach with N neurons. Hence, our approach arrives at a better solution with shorter/processing time and requires less resources as compared to that of the conventional approach.