Combinatorics and Optimization (CO) 769: Topics in Continuous Optimization

Topic for fall 2014:

Optimization for big data


Steve Vavasis

Class times:

12:30-1:30 Monday, Wednesday, Friday.

Experiments carried out nowadays in many branches of science yield huge datasets as their result. Huge datasets are also generated by many social networks and commercial internet sites. The modern paradigm adopted by the machine learning community for analyzing these datasets is to postulate a statistical model that generates the data and that is controlled by a smaller number of unknown parameters. Then the task of analyzing the data is reduced to estimating the parameters. This typically leads to large problems in optimization and numerical linear algebra. In this course, we will consider a few parameter estimation problems found in the recent data mining literature and the algorithms used to solve them.

This course will involve some technically demanding problem sets.

The prerequisites are intentionally kept minimal in order to allow participation by students outside of the Combinatorics and Optimization department.

Prerequisites: Linear algebra; calculus of several variables; previous exposure to linear programming at the level of CO 250.