Abstract: How do we best visualize and understand the neural pathways and activity sites of the brain and their relation to genes and behavior? Building a brain imaging and informatics system that effectively serves the goal of answering this question remains an important scientific computing challenge that requires integration of all numeric and non-numeric approaches to computing. The presentation will discuss various examples ranging from mathematical image processing for computer aided diagnosis to non-numeric symbolic processing in data grid systems for clinical imaging trials. Such methods and tools must be integrated into a future scientific computing system for brain imaging that provides and utilizes a comprehensive array of computing components addressing all major aspects of computational imaging, integrating the necessary mathematics, statistics, numeric-based AI, semantic-based AI, software engineering (user interfaces), and systems engineering (networks).
Numerical integration of large stiff nonlinear systems of ODEs over
long time intervals is a challenging task since the stiffness of the
system makes solving them using standard explicit and implicit
methods computationally expensive. Exponential propagation-type
techniques offer significant advantages for these problems compared
to standard integrators. We will discuss new exponential propagation
iterative (EPI) methods, which are constructed by approximating the
integral form of the solution to a nonlinear autonomous system of
ODEs by an expansion in terms of products between special functions
of matrices and vectors. The matrix functions-vector products are
then approximated using Krylov subspace projections. For problems
where no good preconditioner is available, the EPI integrators can
outperform standard methods since they possess superior stability
properties compared to explicit schemes and offer computational
savings compared to implicit Newton-Krylov integrators by requiring
fewer Arnold iterations per time step. As an application, for which
exponential propagation methods offer computational savings, I will
discuss modeling large-scale plasma behavior using resistive MHD
equations. I will present results of a numerical model which
describes plasma as a time-dependent driven system. The results of
the simulations suggest new structure of plasma configurations that
form in the course of evolution of solar coronal arcades or
laboratory spheromak-type plasmas.
This will be a series of two introductory lectures on the
distribution of closed points on a scheme of finite type
over the integers. Both general properties and important
examples will be discussed, with an emphasis on p-adic
variation for zeta functions over finite fields.
An explicit expression for the L-function evaluator associated to an abelian extension K/k of number fields of degree 2p will be discussed. Instances will be given where this expression can be used to determine pieces of the ideal class group of K that are annihilated by this L-function evaluator.
We present a variational framework for shape optimization
problems that hinges on devising energy decreasing flows based on
shape differential calculus followed by suitable space and time
discretizations (discrete gradient flows). A key ingredient is
the flexibility in choosing appropriate descent directions by
varying the scalar products, used for computation of normal
velocity, on the deformable domain boundary. We discuss
applications to image segmentation, optimal shape design for PDE,
and surface diffusion, along with several simulations exhibiting
large deformations as well as pinching and topological changes in
finite time. This work is joint with E. Baensch, G. Dogan, P.
Morin, and M. Verani.