On the Geometry of Loss Functions 2026-03-08 Why convexity matters more than you think, and what happens when you lose it. math ml optimization