Math/ICES Center of Numerical Analysis Seminar (Spring 2012)

Time and Location: Friday, 3:00-4:00PM. Special time and locations are indicated in color.

If you are interested in meeting a speaker, please contact the host of the speaker.

Here are the links to the past seminars: Fall 2011, Spring 2011, Fall 2010, Spring 2010, Fall 2009

Date

Speaker

Title and Abstract




01/27/2012
Friday
3:00PM-4:00PM
ACE 6.304

Thierry E. Magin              

von Karman Institute for Fluid Dynamics

Kinetic theory derivation of nonequilibrium hydrodynamic models for atmospheric entry plasmas

Atmospheric entry simulations are complex problems due to their intense multiphysics and multiscale nature. The conventional
physico-chemical nonequilibrium models used for these simulations are often derived  "correlation-based" from experiments and stretched out of the validity range for which they have been conceived. We propose to use kinetic theory as a powerful tool to derive macroscopic conservation equations, transport fluxes,  energy exchange terms and chemical production rates for atmospheric entry plasmas. Two approaches are followed: a multiscale Chapman-Enskog perturbative method and Boltzmann-moment system method with Grad closure. The following nonequilibrium effects are studied: mass disparity, electromagnetic field influence, ionization reactions, internal energy excitation, and rarefied gas effects.




02/03/2012
Friday
3:00PM-4:00PM
ACE 6.304

Pradeep Ravikumar

CS, UT Austin

Greedy Algorithms for Structurally Constrained High Dimensional Problems

Modern problems across science and engineering increasingly require high-dimensional models; with more parameters than observations. It is now well understood that statistically reliable inference is still possible under such high-dimensional settings, provided one restricts to constrained subclasses of models with particular low-dimensional structure. Examples include linear regression with sparsity constraints (compressed sensing), estimation of covariance or inverse covariance matrices, sparse principal component analysis, low-rank matrix estimation, and sparse additive non-parametric models. Over the past decade, there has been a strong body of work that have proposed statistical estimators for inferring such structurally constrained high-dimensional models, with strong statistical guarantees. 

In this talk, we consider the computational facet of such estimation: could we provide a general computational scheme to solve any of the convex optimization problems that arise in such high-dimensional inference? We find that such a general computational scheme is indeed possible: specifically, we discuss and analyze a scheme based on a greedy strategy. Our framework not only unifies existing greedy algorithms that have been proposed for such high-dimensional problems by recovering them as special cases but also yields novel ones.







04/02/12
Monday
4:00-5:00PM
RLM 6.104

Gunther Uhlmann

University of California at Irvine and University of Washington