The Community for Technology Leaders
RSS Icon
Subscribe
Issue No.04 - July/August (2002 vol.4)
pp: 22-30
David S. Thompson , Mississippi State University
Jaya Sreevalsan Nair , Mississippi State University
Satya Sridhar Dusi Venkata , Mississippi State University
Raghu K. Machiraju , Ohio State University
Ming Jiang , Ohio State University
Gheorghe Craciun , Ohio State University
ABSTRACT
<p>Exploration of large scientific data sets can be effectively conducted through a feature mining process. At the core of this approach is a binary classification of either individual points followed by an aggregation into contiguous regions or aggregations of points. The resulting collections of points represent regions-of-interest (ROI). The ensuing ROIs are accessed in a progressive manner on a client-server data exploration system. Such a mining effort can significantly reduce client memory requirements for analysis and visualization algorithms. The feature mining efforts we report here are part of a data exploration system called EVITA that is designed for computational fluid dynamics simulations. We describe two paradigms for feature mining and show how we can extract the same feature using the two paradigms. Additionally, we show how to employ a multiscale representation of the data to denoise features that are weak or of small extent. Our methods differ from those reported in the data mining literature in that we exploit the underlying physics of the feature under consideration.</p>
INDEX TERMS
Large data exploration, feature detection, segmentation, feature mining paradigms, feature-preserving wavelet transforms, denoising, computational fluid dynamics
CITATION
David S. Thompson, Jaya Sreevalsan Nair, Satya Sridhar Dusi Venkata, Raghu K. Machiraju, Ming Jiang, Gheorghe Craciun, "Physics-Based Feature Mining for Large Data Exploration", Computing in Science & Engineering, vol.4, no. 4, pp. 22-30, July/August 2002, doi:10.1109/MCISE.2002.1014977
21 ms
(Ver 2.0)

Marketing Automation Platform Marketing Automation Tool