The Cornell Theory Center
The Cornell Theory Center (CTC), directed by Thomas
Coleman, advances Cornell research by providing researchers with a high-performance
computing environment, extensive technical expertise, and support for visualization. CTC
resources include a 160-processor IBM RS/6000 POWER parallel System (SP), a 24-node Intel
NT cluster, and a virtual reality environment called the Visual Insight Zone.
CTC supports a number of interdisciplinary research
projects, several of which apply computer science expertise to problems in other
disciplines. For example, sample projects through CTC's Computational Finance Institute
include exploring new optimization algorithms for the determination of implied volatility
and the solution of the American put pricing model; investigating new computational
procedures for American option problems suitable for large-scale problems based on inverse
problem formulations and new fast methods for non-negative nonlinear least-squares;
developing parallel "scenario optimization" toolkits on top of environments such
as MATLAB for multiple processors and applying automatic differentiation technology; and
adapting new algorithms suitable for large problems and parallel computing for norm and
down-side norm minimization applied to portfolio analysis and value-at-risk calculations.
CTC's Parallel Processing Resource for Biomedical
Scientists focuses on pushing algorithm development and software advances to lay the basis
for further advances in biomedical science. The Parallel Resource is aimed primarily at
the protein folding problem and structure-based drug design, but computer scientists have
also worked with imaging problems. Projects have included image segmentation for lung
nodule detection; high-resolution and quantitative ultrasonic imaging; rational design of
MRI contrast enhancement agents; and the development of a quantitative theory for physical
phenomena that affect MRI microscopy and the numerical modeling of these phenomena.
Another interdisciplinary project, Crack Propagation
on Teraflop Computers, focuses on the design of algorithms and systems to support the
numerical simulation of crack propagation problems. The focus is on 3-dimensional,
time-dependent fracture simulations using unstructured, adaptive grids on the SP and on
EARTH-MANNA, a machine based on a fine-grain, multi-headed program execution model. The
effort involves three major tasks: designing and implementing 3-D adaptive, parallel mesh
generators producing meshes of provably good quality; developing and implementing
restructuring compiler technology to support the automatic generation of efficient
parallel code, starting with equations that arise in the simulation; and evaluating the
adequacy of existing thread generation and load balancing mechanisms on the parallel
platforms and reimplementing the mechanisms for crack propagation studies on parallel
computers.
A new and growing interdisciplinary initiative
centers around genomics. Premised on the idea that the ability to interpret genomic data
will be the chief enabling technology that will lead to the next major breakthroughs in
understanding the relationship between genomic structure and biological function, the
Cornell Genomics Initiative has a strong computational genomics and bioinformatics
component. For example, one group is trying to see if inexpensive parallelization
techniques can be used to increase the speed of ACEDB genome database query responses.
ACEDB is the pre-eminent database management system for agricultural, human, and
model-biological-system genome projects world-wide.
A number of faculty integrate the use of CTC
high-performance computing resources into their courses. In CS, this has occurred in
courses such as CS 417/418 (Computer Graphics and a Computer Graphics Lab, which cover the
basics of modeling, rendering, and animation), CS 522 (Computational Tools and Methods for
Finance), and CS 612 (Software Design for High Performance Architectures). |