Structural Inferences from Massive Datasets

Abstract

High-level understanding of data must involve the interplay between substantial prior knowledge with geometric and statistical techniques. Our approach emphasizes the recovery of basic structural elements and their interaction patterns in order to summarize and draw inferences about the significant features contained in the data. As a testbed for modeling how scientists analyze and extract knowledge of structure morphogenesis from data , we examine the datasets obtained from numerical simulation of turbulence. We describe a program that automatically extracts 3D structures, classifies them geometrically, and analyzes their spatial and temporal coherence. Our program is constructed by mixing and matching the aggregate, classify, and re-describe operators of the spatial aggregation language. The research is a continuation of the effort to investigate the role of imagistic reasoning in human thinking.