Weight annealing

Abstract

Simulated Annealing has become a standard optimization technique for a wide variety of problems: starting at a random configuration and performing a sequence of moves, the system is optimized using a control parameter which partially allows for accepting a deterioration and therefore for climbing over barriers in the energy landscape. Our approach, Weight Annealing, changes the energy landscape by assigning variable weights to the single parts of the proposed problem. We describe the philosophies behind these algorithms and present results for the Traveling Salesman Problem and the Sherrington–Kirkpatrick-model for spin glasses.