Title: Machine Learning Seminar(2016.10.19)
Author: CMAC
Date: 2016-10-18 (17:28)

Machine Learning Seminar

Date/Time: Oct. 19th, Wed., 11:00 ~ ?AM

Location: Office of Prof. Choe(#405), Yonsei University

Speaker: Prof. Sang Uhn Yoon

Affiliation: Sungkyunkwan University

Topic: First-order methods for structured nonsmooth optimization


We consider the structured nonsmooth optimization problem whose objective function is the sum of a smooth function and a proper (convex) lowersemicontinuous function. The special cases of this structured nonsmooth optimization problem are bounded constraint optimization problem, L1-regularized linear least squares problem, L1-regularized logistic regression problem, support vector machines, total variation regularized convex minimization, nuclear norm regularized problem, sparse covariance selection problem. We introduce three optimization methods, coordinate gradient method, incremental gradient method, and splitting method, for solving structured nonsmooth optimization problems. These methods have also attracted much interest in bigdata.