Lawrence Berkeley National Laboratory

One fundamental element of scientific inquiry is discovering relationships, particularly the interactions between different variables in observed or simulated phenomena. Building upon our prior work in the field of Query-Driven Visualization, where visual data analysis processing is focused on subsets of large data deemed to be "scientifically interesting," this new work focuses on a novel knowledge discovery capability suitable for use with petascale class datasets. It enables visual presentation of the presence or absence of relationships (correlations) between variables in data subsets produced by Query-Driven methodologies. This technique holds great potential for enabling knowledge discovery from large and complex datasets currently emerging from SciDAC and INCITE projects. It is sufficiently generally to be applicable to any time of complex, time-varying, multivariate data from structured, unstructured or adaptive grids.