SAMSI
Distinguished Lectures 2003-2004
MCNC-RDI
Auditorium
4:30-5:30pm (followed by a reception in the lobby)
Tuesday - October 7, 2003
Margaret
Wright
New York University, Computer Science
Department & Courant Institute of Mathematical Sciences
"Direct Search
Models: The Sound and The Fuss"
Abstract: Direct search methods, which optimize without using derivatives, constitute a fascinating chapter, still being written, in the annals of optimization. Their history includes an initial heyday in the 1960s, a fall from grace in the 1970s and 80s, and a resurgence dating from the mid-1990s. Today's research on these methods has highlighted several interesting---in some cases, contentious---issues, ranging from the nature of convergence proofs and the associated assumptions to the proper role for direct search methods in modern optimization practice. We shall consider a selection of these issues, their current status, and their implications. And since this talk is being given at SAMSI, the speaker has high hopes for an improvisational segment devoted to connections between direct search methods and statistical sampling.
Tuesday - November 4, 2003
Jerome Friedman
Stanford University, Department of Statistics & Stanford Linear
Accelerator Center
"Importance Sampling: An Alternative View of
Ensemble Learning"
Abstract: Learning a function of many arguments is viewed from the perspective of high-dimensional numerical integration. It is shown that many of the popular ensemble learning methods can be cast in this framework. In particular, bagging, boosting, and Bayesian model averaging are seen to correspond to Monte Carlo integration methods each based on different importance sampling strategies. This interpretation explains some of their properties and suggests modifications to them that can improve their accuracy and especially their computational performance.
* Joint work with Bogdan Popescu
Tuesday - March 2, 2004
Jonathan Chapman
Oxford University, Mathematical Institute & Oxford Centre for
Industrial and Applied Mathematics
"A hierarchy of models for type-II
superconductors"
Abstract: One of the interesting aspects of superconductivity is the variety of models available to describe the phenomenon at different lengthscales, ranging from the microscopic theory of Bardeen, Cooper & Schreiffer through the mesoscopic theories of London and Ginzburg & Landau, to the macroscopic Critical State theories such as the Bean model. The talk will explore the relationship between these different models by examining suitable asymptotic limits.
The basic building block in deriving this hierarchy is the superconducting vortex, which is a thin core of nonsuperconducting material circled by a superconducting electric current. Similar line singularities are found in other systems, for example, line vortices in an inviscid fluid, or Volterra dislocations in an elastic crystal. This raises the interesting question of whether analogous hierarchies of models exist in each of these other systems, and whether similar connections can be established between them.
Tuesday - April 6, 2004
Thomas G. Kurtz
University of Wisconsin-Madison, Department of Mathematics &
Department of Statistics
"Particle representations of continuum
models"
Abstract: Many stochastic and deterministic models are derived as the continuum limit of a discrete stochastic system as the size of the system tends to infinity. Discrete "particles" are each assigned a small mass and the limiting "mass distribution," typically characterized as a solution of a deterministic or stochastic partial differential equation, gives the desired model. A number of examples will be described in which keeping the discrete particles in the limit provides a useful tool for justifying the limit, analyzing the limiting model, and relating data to the limiting model. Possible examples include derivation of fluid models for internet protocols, many-server queueing approximations, models of stock prices set by infinitely many competing traders, consistency of numerical schemes for filtering equations, and models in population genetics; however, the speaker promises not to try to discuss all of these.
Directions to MCNC, Research Triangle Park
Exit the Durham Freeway
(Route 147) onto Cornwallis Road going West.
MCNC is on
the right-hand side of Cornwallis Road, 0.2 miles after the first light.
The
auditorium is in the building on the left.
Entire site © 2001-2004, Statistical and Applied Mathematical Sciences Institute. All Rights Reserved.