FoCM 2014 conference
Workshop B3 - Continuous Optimization
December 15, 14:30 ~ 15:00 - Room A21
Randomized Methods for Zeroth-Order Optimization
Alexander Rakhlin
University of Pennsylvania, USA - rakhlin@gmail.com
We present several methods for zeroth order stochastic convex optimization and analyze their complexity. The proposed algorithms are based on random walks on convex bodies. We find that such methods can deal with the noisy information in a more stable manner.
Joint work with A. Belloni (Duke University, USA), T. Liang (University of Pennsylvania, USA) and H. Narayanan (University of Washington, USA).