Applying Self-Organizing Kohonen Maps to the Traveling Salesman Problem

This paper discusses solving the traveling salesman problem using Kohonen maps. We investigate the application of Self-Organizing Maps (SOM) to the Traveling Salesman Problem (TSP), develop Central Processing Unit (CPU)- and Graphics Processing Unit (GPU)-based implementations of the proposed approach, and evaluate their efficiency in SOM hyperparameter tuning using a multistage dimensionality reduction framework and Bayesian optimization. Bayesian optimization based on sequential trial selection was used to select the parameters. A new SOM configuration was defined for each trial, after which the map was trained, and the objective function value was calculated. This approach differs from exhaustive searches in that subsequent trials are selected based on the results already obtained.