The Most Likely Steady State for Large Numbers of Stochastic Traveling Salesmen
Robert Geist, Robert Reynolds · 2021
A class of continuous time, ergodic Markov processes having extremely large state space is considered, and a technique for numerical determination of high-probability components of the steady-state solution vector is proposed. The technique first identifies the steady-state solution as a maximum entropy Gibbs measure , and then maps the associated energy function onto a Hopfield neural network . A numerical iteration scheme based on an unresolved conjecture due to Hillam is used to complete the computation. An application of this technique to computer graphics is offered in the form of a new algorithm for digital halftone resolution.