To verify that the solution is optimal, the algorithm checks whether the candidate is a global maximum by comparing its value to every other feasible point.

When optimizing neural networks, researchers must be careful not to confuse local peaks with global maximums in the loss landscape.