Graduate Student Seminar

Speaker 1: Sumaya Alzuhairy

Title

Efficient multilevel methods for optimal control of elliptic equations with stochastic coefficients

Abstract

A common strategy for solving optimal control of stochastic PDEs relies on stochastic collocation, which reduces the problem to multiple solves of optimal control problems constrained by deterministic PDEs.

In this work we investigate an alternative approach where we use a stochastic Galerkin formulation and discretization of the PDE prior to solving the optimal control problem. Ultimately this requires solving a potentially very large linear system, which we then solve using specially designed multilevel algorithms.

This is based on a joint work with Dr. Draganescu and Dr. Sousedik.

Speaker 2: Reetam Majumdar

Title

A Comparison of Logistic Regression and Naïve Bayes for Classification

Abstract

Logistic regression and Naïve Bayes are two commonly used techniques in classification problems; the former is a Generalized Linear Model and an analogue to OLS regression, and the latter forms a fast and easy to execute family of probabilistic classifiers. We introduce both methods and discuss three papers comparing their performances – Rish (2001), Ng and Jordan (2002), and Xue and Titterington (2008). As a follow up, simulations are run on a dataset used in one of the papers to see how both perform, and if any criterion can be established for choosing one method over the other.