Construction of effective noise for TSP

We investigate searching ability finding local minima and the global minimum in a given traveling salesman problem (TSP) by using the Hopfield neural network with stochastic noise sources composed of various time series. Firstly we tune the network up by changing parameters of a 2-state Gilbert model noise source whose time series looks like an intermittency chaos. As a result, the solving ability cannot be improved drastically by changing such parameters. Secondly, we propose two noise sources; a noise generated by switching two different periodic motions, a noise generated by m-state Gilbert model with different periodic motions. From the numerical experiments, we can conclude that the periodic behavior and its stochastical. switching is rather essential as an effective noise for TSP.