There has been a long history of using neural networks for combinatorial optimization and constraint satisfaction problems. Symmetric Hopfield networks and similar approaches use steepest descent dynamics. and they always converge to the closest local minimum of the energy landscape. For finding global minima additional parameter-sensitive techniques are used. https://www.bekindtopets.com/deal-alert-NHL-Winnipeg-Jets-Distressed-iPhone-13-Pro-Max-MagSafe-Case-p96268-great-save/