Higher-Order Multi-Variate Statistics for Scientific Data Analysis

Explore higher-order multi-variate statistics for scientific data analysis in this seminar by Hemanth Kolla from Sandia National Laboratories. Delve into the importance of analyzing multi-variate non-Gaussian statistical processes in scientific phenomena, going beyond traditional correlations and covariance. Learn about the application of higher-order statistics like coskewness and cokurtosis in scientific data analysis, with a focus on rare event detection and dimensionality reduction for stiff dynamical systems. Discover connections to Independent Component Analysis and symmetric tensor decomposition. Gain insights into the development of low overhead, scalable, and parallelizable algorithms for computing and factorizing the cokurtosis tensor, designed for in situ application in large simulations. Understand the intersection of high-performance scientific computing and statistical learning, including topics such as tensor decompositions, efficient forward propagation of parametric uncertainty in computational mechanics, and algorithm-based fault tolerance for HPC.

DSI | Higher-Order Multi-Variate Statistics for Scientific Data Analysis