Visual Analysis of Simulation Uncertainty Using Cost-Effective Sampling
Annie Preston, Yiran Li, Franz Sauer, and Kwan-Liu Ma
IEEE Symposium on Large Data Analysis and Visualization (LDAV) 2018, 2018
Abstract
Studying large, complex simulations entails understanding their uncertainties. However, visualization tools that rapidly quantify simulation uncertainty may require precise tuning, give limited information, or struggle to disentangle uncertainty sources. We propose a fast, scalable regression-based approach that uses bootstrapping on small samples of simulation data to model the effect of uncertainty from discreteness. We test the approach on three types of simulations with unique sources of uncertainty: particles (dark matter), ensembles (ocean), and discretized flows (traffic). We create a visualization tool to facilitate this modeling, showing training data and predictions in real time. Scientists, who need to provide only modest supervision, can use our tool to quickly understand how initial conditions and parameterizations affect observable quantities, their uncertainties, and their agreement with experimental data. We show that our tool offers a speedup of several orders of magnitude over comparable uncertainty calculation approaches.
BibTex
@INPROCEEDINGS{preston2018visual,
author={Preston, Annie and Li, Yiran and Sauer, Franz and Ma, Kwan-Liu},
booktitle={2018 IEEE 8th Symposium on Large Data Analysis and Visualization (LDAV)},
title={Visual Analysis of Simulation Uncertainty Using Cost-Effective Sampling},
year={2018},
pages={1-11},
doi={10.1109/LDAV.2018.8739182}}