Joint Seminar, Lefschetz Center for Dynamical Systems and
The Center for Fluid Mechanics Seminar
Abstract: Electrospinning is a process which creates sub micron sized fibers from fluid forced through a millimeter sized nozzle by an electric field. We present evidence from theory and experiments that the essential mechanism underlying this process is a rapidly whipping fluid jet. An asymptotic approximation of the equations of electrohydrodynamics is developed so that quantitative comparisons with experiments can be carried out. The onset threshold for electrospinning is demonstrated to quantitatively agree with the onset of a whipping instability. Scaling laws for the whipping frequency are derived. It is demonstrated that quantitative features of the instability depend sensitively on the shape and material properties of the nozzle.
Special Brown University Mathematics Department Seminar
Brown Applied Mathematics Pattern Theory and Vision Seminar
Abstract: In the early 1990s Dragon Systems embarked on an ambitious research program that aimed to explore the question of whether it would be possible to use large vocabulary speech recognition to assist in determining who the speakers in a phone conversation might be and what topic they might be speaking about. This research was facilitated by the collection of a corpus of recorded (and transcribed) phone conversations (the "Switchboardcorpus") involving speakers who had been asked to converse on a given topic. The originsal topic id problem posed by this data has not been essentially solved (but only because the problem had been made artificially easy by design), while the speaker id problem remains a challenging one. Later on in the 90s, subtler problems of topical content identification and story segmentation were posed in the government's TDT (topic detection and tracking) program. In this talk I will review some of the approaches that we have taken to these problems and give some indication of how well the existing technology actually functions. An important theme of the talk will be the connections that exist between the methods used in speech recognition and those that have been applied to the speaker and topic problems that I discuss here. Although there has been some success achieved so far in dealing with these problems, the methods being applied are still quite primitive. New ideas will be needed!
Scientific Computing Seminar
Old Dominion University | |
Abstract: Inexact subspace preconditioning techniques of Schur and Schwarz varieties have revolutionized the parallel numerical analysis of partial differential equations through domain decomposition, leading to efficient employment of thousands of processors in practical applications. While originally motivated by results from analysis, these paradigms can be generalized beyond any present supporting theory at the discrete algebraic level to applications in multidisciplinary simulation and in optimization. In this talk, we review the algorithmic basis and architectural motivation for partitioned solvers generally, furnish a high-end example from computational aerodynamics in some detail, and briefly discuss extensions, including the new Lagrange-Newton-Krylov-Schur-Schwarz family of optimization algorithms.
PDE Seminar
University of Massachusetts, Amherst | |
<--- 2000 Index