Reminder: local optima were never a problem in machine learning, where the goal is generalization, not fitting the training data perfectly.

→ View original post on X — @pmddomingos