Exploration and Analysis of Image-base Simulation Ensembles

Abstract

Scientists run simulation ensembles to study the behavior of a phenomenon using varying initial conditions or input parameters. However, the I/O bottlenecks hinder performing large-scale multidimensional simulations. In situ visualization approaches address the variability of I/O performance by processing output data during simulation time and saving predetermined visualizations in image databases. This poster proposes a visual analytics approach to exploring and analyzing image-based simulation ensembles, taking advantage of semantic interaction, feature extraction, and deep learning techniques. Our approach uses deep learning and local feature techniques to learn image features and pass them along with the input parameters to the visualization pipeline for in-depth exploration and analysis of parameter and ensemble spaces simultaneously.