PNNL-13501 - Pacific Northwest National Laboratory

Study Control Number: PN99078/1406
Visualization and Active Guidance of Scientific Computations
Pak Chung Wong
We developed the concept of a data signature, which captures the essence of a large, multidimensional scientific data set
in a compact format, and used it to conduct analysis as if using the original. Our results have shown that the technology
can be applied to a wide variety of computational sciences and engineering studies—such as climate, combustion,
imaging, and computational fluid dynamics.
Project Description
The purpose of this project is to develop the concept of a
data signature that captures the essence of a multidimensional
scientific data set in a compact format, and
use it to conduct analysis as if using the original. A data
signature, because of its compact design, always requires
fewer computational resources than its original. We
developed two novel approaches based on feature
extraction and wavelet transforms for creating data
signatures of very large time-dependent data sets. We
successfully developed our first signature design (featurebased)
on multiple parallel and desktop computer
platforms. This system was applied to different
simulation studies, including global climate and
combustion. The experimental results were documented
in two peer-reviewed journal and conference papers.
Introduction
Today, as data sets used in computations grow in size and
complexity, the technologies developed over the years to
manage scientific data sets have become less efficient and
effective. Many frequently used operations, such as
Eigenvector computation, could quickly exhaust our
desktop workstations once the data size reached certain
limits. While building new machines with more resources
might conquer the data size problems, the complexity of
today’s computations requires a new breed of projection
techniques to support analysis of the data and verification
of the results. Our solution is the data signature, that
captures the essence of a scientific data set in a compact
format and is then used to conduct analysis as if using the
original.
A data signature can be described as a mathematical data
vector that captures the essence of a large data set in a
fraction of its original size. It is designed to characterize
a portion of a data set, such as an individual time frame of
a scientific simulation. These signatures enable us to
conduct analysis at a higher level of abstraction and yet
still reflect the intended results as if using the original
data.
Approach
To date, we have investigated designing signatures for
scalar fields, tensor fields, and a combination of them
with multiple parameters. The construction of a data
signature is based on one or more of the following
features and approaches:
• velocity gradient tensors
• critical points and their eigenvalues
• orthogonal and non-orthogonal edges
• covariance matrices
• intensity histograms
• content segmentation
• conditional probability.
The selected features often depend on data type.
Information such as velocity gradient and critical points is
suitable for vector and tensor field data. Edge detection,
covariance, and histogram can be applied to scalar data.
For large scientific simulations, we sometimes rely on a
multi-resolution approach to determine the desirable size
of the signatures.
Results and Accomplishments
In theory, a data signature can never be as descriptive as
its full-size original. In practice, however, a well-defined
data signature can be as good or better from many
perspectives because it brings out specific details and
eliminates less important information from consideration.
The concept is useful when we study the characteristics of
scientific simulations, such as global climate modeling.
Computer Science and Information Technology 169