Matrix-Free Preconditioning in Online Learning
Ashok Cutkosky · Tamas Sarlos

Thu Jun 13th 09:00 -- 09:20 AM @ Room 102

We provide an online convex optimization algorithm with regret that interpolates between the regret of an algorithm using an optimal preconditioning matrix and one using a diagonal preconditioning matrix. Our regret bound is never worse than that obtained by diagonal preconditioning, and in certain setting even surpasses that of algorithms with full-matrix preconditioning. Importantly, our algorithm runs in the same time and space complexity as online gradient descent. Along the way we incorporate new techniques that mildly streamline and improve logarithmic factors in prior regret analyses. We conclude by benchmarking our algorithm on synthetic data and deep learning tasks.

Author Information

Ashok Cutkosky (Google)
Tamas Sarlos (Google)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors