Robust Accelerated Gradient Methods

Monday, May 6, 2019 - 1:25pm - 2:25pm
Lind 305
Mert Gurbuzbalaban (Rutgers, The State University Of New Jersey)
We study the problem of minimizing a strongly convex and smooth function when we have noisy estimates of its gradient. We propose a novel multistage accelerated algorithm that is universally optimal in the sense that it achieves the optimal rate both in the deterministic and stochastic case and operates without knowledge of noise characteristics. The algorithm consists of stages that use a stochastic version of Nesterov's accelerated algorithm with a specific restart and parameters selected to achieve the fastest reduction in the bias-variance terms in the convergence rate bounds.

Mert Gürbüzbalaban is an assistant professor at Rutgers University. Previously, he was a postdoctoral associate at the Laboratory for Information and Decision Systems (LIDS) at MIT. He is broadly interested in optimization and computational science driven by applications in large-scale information and decision systems. He received his B.Sc. degrees in Electrical Engineering and Mathematics as a valedictorian from Boğaziçi University, Istanbul, Turkey, the “Diplôme d’ingénieur” degree from École Polytechnique, France, and the M.S. and Ph.D. degrees in (Applied) Mathematics from the Courant Institute of Mathematical Sciences, New York University.

Dr. Gürbüzbalaban received the Kurt Friedrichs Prize (given by the Courant Institute of New York University for an outstanding thesis) in 2013, Bronze Medal in the École Polytechnique Scientific Project Competition in 2006, the Nadir Orhan Bengisu Award (given by the electrical-electronics engineering department of Boğaziçi University to the best graduating undergraduate student) in 2005 and the Bülent Kerim Altay Award from the Electrical-Electronics Engineering Department of Middle East Technical University in 2001. He received funding from a variety of sources including multiple programs at the U.S. National Science Foundation.