Seminars and Colloquia by Series

Thursday, April 16, 2009 - 15:00 , Location: Skiles 269 , Vladimir I. Koltchinskii , School of Mathematics, Georgia Tech , Organizer: Heinrich Matzinger
In binary classification problems, the goal is to estimate a function g*:S -> {-1,1} minimizing the generalization error (or the risk) L(g):=P{(x,y):y \neq g(x)}, where P is a probability distribution in S x {-1,1}. The distribution P is unknown and estimators \hat g of g* are based on a finite number of independent random couples (X_j,Y_j) sampled from P. It is of interest to have upper bounds on the excess risk {\cal E}(\hat g):=L(\hat g) - L(g_{\ast}) of such estimators that hold with a high probability and that take into account reasonable measures of complexity of classification problems (such as, for instance, VC-dimension). We will discuss several approaches (both old and new) to excess risk bounds in classification, including some recent results on excess risk in so called active learning.
Wednesday, April 15, 2009 - 13:30 , Location: ISyE Executive Classroom , Luke Postle , School of Mathematics/ACO, Georgia Tech , Organizer: Annette Rohrs
Grotzsch's Theorem states that every triangle-free planar graph is 3-colorable. Thomassen conjectured that every triangle-free planar graph has exponentially many distinct 3-colorings. He proved that it has at least 2^{n^{1/12}/20000} distinct 3-colorings where n is the number of vertices. We show that it has at least 2^{\sqrt{n/600}} distinct 3-colorings. Joint work with Arash Asadi and Robin Thomas.
Wednesday, April 15, 2009 - 12:00 , Location: Skiles 255 , Sung Ha Kang , School of Mathematics, Georgia Tech , Organizer:
This talk will focus on mathematical approaches using PDE and variational models for image processing. I will discuss general problems arising from image reconstructions and segmentation, starting from Total Variation minimization (TV) model and Mumford-Shah segmentation model, and present new models from various developments. Two main topics will be on variational approaches to image reconstruction and multi-phase segmentation. Many challenges and various problems will be presented with some numerical results.
Wednesday, April 15, 2009 - 11:00 , Location: Skiles 255 , Igor Belykh , University of Georgia , Organizer:
Series: PDE Seminar
Tuesday, April 14, 2009 - 15:05 , Location: Skiles 255 , Joy Ko , Brown University, Providence , Organizer:
I will talk about the highlights of a collaborative and multidisciplinary program investigating qualitative features of steady water waves with vorticity in two dimensions. Computational and analytical results together with data from the oceanographic community have resulted in strong evidence that key qualitative features such as amplitude, depth, streamline shape and pressure profile can be fundamentally affected by the presence of vorticity. Systematic studies of constant vorticity and shear vorticity functions will be presented.
Monday, April 13, 2009 - 16:30 , Location: Skiles 255 , Konstantin Mischaikow , Rutgers University , Organizer: Haomin Zhou
I will discuss new computational tools based on topological methods that extracts coarse, but rigorous, combinatorial descriptions of global dynamics of multiparameter nonlinear systems. These techniques are motivated by the fact that these systems can produce an wide variety of complicated dynamics that vary dramatically as a function of changes in the nonlinearities and the following associated challenges which we claim can, at least in part, be addressed. 1. In many applications there are models for the dynamics, but specific parameters are unknown or not directly computable. To identify the parameters one needs to be able to match dynamics produced by the model against that which is observed experimentally. 2. Experimental measurements are often too crude to identify classical dynamical structures such as fixed points or periodic orbits, let alone more the complicated structures associated with chaotic dynamics. 3. Often the models themselves are based on nonlinearities that a chosen because of heuristic arguments or because they are easy to fit to data, as opposed to being derived from first principles. Thus, one wants to be able to separate the scientific conclusions from the particular nonlinearities of the equations. To make the above mentioned comments concrete I will describe the techniques in the context of a simple model arising in population biology.
Series: Other Talks
Monday, April 13, 2009 - 16:30 , Location: Skiles 269 , Jozsef Solymosi , Math, UBC , Organizer: Prasad Tetali
An old conjecture of Erdos and Szemeredi states that if A is a finite set of integers then the sum-set or the product-set should be large. The sum-set of A is A + A={a+b | a,b \in A\}, and the product set is defined in a similar way, A*A={ab | a,b \in A}. Erdos and Szemeredi conjectured that the sum-set or the product set is almost quadratic in |A|, i.e. max(|A+A|,|A*A|)> c|A|^{2-\epsilon}. In this talk we review some recent developments and problems related to the conjecture.
Monday, April 13, 2009 - 14:00 , Location: Skiles 269 , Roman Golovko , USC , Organizer: John Etnyre
We will define the sutured version of embedded contact homology for sutured contact 3-manifolds. After that, we will show that the sutured version of embedded contact homology of S^1\times D^2, equipped with 2n sutures of integral or infinite slope on the boundary, coincides with the sutured Floer homology.
Monday, April 13, 2009 - 14:00 , Location: Skiles 255 , Doron Lubinsky , School of Mathematics, Georgia Tech , Organizer: Plamen Iliev
It turns out that the sinc kernel is not the only kernel that arises as a universality limit coming from random matrices associated with measures with compact support. Any reproducing kernel for a de Branges space that is equivalent to a Paley-Winer space may arise. We discuss this and some other results involving de Branges spaces, universality, and orthogonal polynomials.
Monday, April 13, 2009 - 13:00 , Location: Skiles 269 , Uli Walther , Purdue University , Organizer: Stavros Garoufalidis
Starting with some classical hypergeometric functions, we explain how to derive their classical univariate differential equations. A severe change of coordinates transforms this ODE into a system of PDE's that has nice geometric aspects. This type of system, called A-hypergeometric, was introduced by Gelfand, Graev, Kapranov and Zelevinsky in about 1985. We explain some basic questions regarding these systems. These are addressed through homology, combinatorics, and toric geometry.