What is the difference between global and local minima in supervised learning optimization?

Global minima is the lowest possible loss, while local minima are smaller valleys that are not the absolute best.

In optimization, the global minimum represents the best possible solution with the lowest loss. Local minima are points where loss is lower than nearby points but not the lowest overall. In deep learning, due to high dimensionality, local minima are less problematic than saddle points. SGD and its variants help navigate these landscapes effectively.