Date
Tue, 28 Apr 2015
Time
14:00 - 15:00
Location
L3
Speaker
Kimon Fountoulakis
Organisation
University of Edinburgh
Support vector machines and signal reconstruction problems have initiated a resurgence of optimization methods with inexpensive iterations, namely first-order methods. The efficiency of first-order methods has been shown for several well conditioned instances. However, their practical convergence might be slow on ill-conditioned/pathological instances.
 
In this talk we will consider Newton-type methods, which aim to exploit the trade-off between inexpensive iterations and robustness. Two methods will be presented, a robust block coordinate descent method and a primal-dual Newton conjugate gradients method.  We will discuss theoretical properties of the methods and we will present numerical experiments on large scale l1-regularized logistic regression and total variation problems.
Please contact us with feedback and comments about this page. Last updated on 03 Apr 2022 01:32.