|
| |
Introduction
New Computational Challenges (NCC) focuses on research and tools to discover,
model, simulate, display, and understand complex systems or complicated
phenomena; to control resources or deal with massive volumes of data in real
time, particularly distributed resources or data; to represent, predict, and
design complex systems; and to understand their behaviors. NCC builds on the
success, but broadens the scope, of prior NSF efforts such as the Grand
Challenge initiatives.
NCC aims to enable wide scientific collaboration and effective management of
complex systems. This will require significant advances in hardware and software
to handle multiple representations, scales, and structures; to enable
distributed collaboration among disparate communities; and to facilitate
real-time interactions and control.
Many phenomena are too complicated to understand in detail from simple
observation or by reduction to isolated components and often require the
coupling of disciplinary scientists and engineers and those involved in enabling
methods and technologies in order to produce new ways to approach previously
intractable problems. The very structure of the problem --- its mathematical,
logical, or computational form --- may change as scale, level of resolution, or
granularity changes. Many important problems require multiple data types,
qualitative information, feedback during the computation to steer it, and a
variety of numerical and symbolic computations. Advances in raw computing power
have outpaced the effectiveness of existing tools and the degree to which they
will scale to large numbers of distributed systems. The development of
meaningful simulations that combine disparately structured models into new types
of simulations is critical. While understanding complex phenomena is obviously
important, predicting their behavior and potentially controlling or changing it,
and doing so in real time, alter the fundamental nature of the problem and
introduce enormous challenges across a broad spectrum of science and engineering
research.
Research Emphases for FY 1998
As noted in the introduction, many scientific and engineering problems are
encompassed by new challenges in computation. For Fiscal Year 1998, NCC will
emphasize only two of these:
scientific and engineering problems involving interactions between phenomena at
different scales or structures, and
problems requiring a dynamic interplay between computations and data.
Proposals that address these two problems or that use them to motivate advances
in enabling computational technologies are welcome. In subsequent years, the
focus of NCC may be broadened to include other computational challenges in
addition to the two chosen for emphasis in FY 1998.
Problems of Scale and Structure
Problems involving multiple scales in space or time occur throughout engineering
and science. Examples include inferring macroscopic properties of a material
from its microstructure; turbulence, which plays a critical role in fluid flows
as varied as mixing of fuel and air in combustion engines, airflow around an
airplane, and blood flow in the heart; scaling of flow in porous media from the
pore level to the field level, which has important applications to oil recovery
and environmental issues; and fluid circulation in the oceans and the
atmosphere. The brain, a dynamic, highly-connected, multi-level organization,
involves both scale and structure. An overlapping set of complex computational
problems are those concerning phenomena that arise from interactions among large
numbers of relatively simple objects or elements. Examples include the complex
perceptual and cognitive phenomena that arise from interactions among neurons in
the brain; the behavior of the immune system in responding to antigens; social
behaviors in animals ranging from insects to humans; human economic and social
activities; and, the operation of distribution networks such as power grids and
communication systems.
Interplay Between Computations and Data
Better understanding of complex phenomena now requires a dynamic interplay
between computations and data, often in real time. Most simulations are entirely
initial-value in style: guess at a start, compute, see what happens, then change
the guess. Simulations that could adapt to intermediate results or changing data
would greatly reduce the number of iterations. In addition, some problems
require this adaptive interplay for effective solution. These include
command-control problems such as air traffic control, dispatch systems, radar
and sonar identification, and other recognize-and-respond problems. Resource
management and process control problems, especially with time constraints, are
also of this kind.
Data-mining problems are of a different nature. Here the idea is to discover
"unusual" items in a large dataset. Examples arise in seismology, high-energy
physics, astronomy, credit card fraud, and management and protection of
networked resources such as databases or computers.
Another kind of problem is combining different kinds of data. There are
difficulties in validating data, assessing the effects of individual errors and
their combinations, and in representing and visualizing data; practical methods
for a multiplicity of large-scale datasets are needed.
Understanding of complex phenomena often depends on mapping different kinds of
data against each other. Examples include tracking any time evolution or spatial
evolution of phenomena against a spatial database (GIS, satellite and other map
data), such as agricultural data, erosion and floods, epidemics, and other
ecological/environmental phenomena; and mapping measurements of a behavior
against measurements of physiological change, e.g., speech or vision against
brain activity.
The examples given in the preceding paragraphs are meant to be illustrative and
not limiting.
| |
|
