ITR: Associative Mining of Large Datasets

Rutgers University New Brunswick, New Brunswick NJ

Investigators

Abstract

The goal of this project is to develop a methodology and set of prototype tools to enable "associative" mining of very large scientific data sets. The researchers will use content, such as solution features, patterns, and shapes, to examine the data and retrieve required information. Unlike other approaches that use index-based coordinates, this lets scientists answer the kinds of questions that they typically ask - such as "Have I seen this evolution before?" and "Is it similar to any experimental observations?" The tools developed by this project will operate on distributed time-varying data and will act as templates for other methods. Specific technical objectives include developing distributed multi-resolution techniques for cataloging interesting phenomena and searching both run-time and archived data for interesting phenomena. The research will specifically target two domains that are representative of other scientific areas and have a pressing need for scientific mining tools: fluid dynamics (large-scale, high-accuracy Direct Numerical Simulation of compressible turbulence) and oceanography (comparison of simulation data with acoustic observations of hydrothermal plumes).