Princeton Prof Reviews MIT Nonconvex Optimization Paper: Prize Remains Open

HazanPrinceton · x · 2026-07-30

Princeton professor Elad Hazan reviewed a new MIT paper on nonconvex optimization. The paper proposes a black-box approach reducing nonconvex optimization to static regret minimization in online convex optimization (OCO).

Hazan noted that while it is a nice paper, it does not solve his open prize question regarding achieving an AdaGrad speedup in nonconvex optimization. The paper's bound is based on an "SGD rate + regret" approach, where a 2-expert Hedge already provides an SGD-safe guarantee on their surrogate. Thus, his prize remains unclaimed.

Original post →

More from Research

Research channel →