Large scale traveling salesman problem via neural network divide and conquer

Among the early motivations for research in neural networks were works suggesting that they would show promise for combinatorial optimization problems such as the Traveling Salesman Problem. These hopes appeared to be disappointed by over a decade of disappointing results, due to scaling problems. However, these problems can be overcome, by application of divide-and-conquer strategies. Our results demonstrate that neural networks are capable of solving problems in the quarter-million city range, with reasonable computational costs. Tour quality for this size problem remains poor, but the use of standard crossover removal algorithms should bring quality into an acceptable range for many applications.