Ensemble-Aware Uncertainty VisualizationSpeaker: Jibonananda Sanyal – Oak Ridge, TN, United States
Topic(s): Information Systems, Search, Information Retrieval, Database Systems, Data Mining, Data Science
Understanding uncertainty in scientific simulations is fundamental in gaining reliable insight into a scientific process. In simulation and modeling, ensembles approaches are recognized as a statistically valid method to capture modeling inexactness thereby yielding higher confidence in the results. Yet, approaches that go beyond traditional workflow management and treat an ensemble simulation run as one large experiment is lacking. Further, the support for analytics and visualization across the entire ensemble dataset is particularly problematic as current datasets tend towards petabytes or more and execute across 1000s of compute nodes, and often in batches.
The challenge is particularly acute for researchers who use leadership class computing infrastructure. Deriving insight from not just single model output but across 10s, sometimes even 1000s of model runs across many thousands of nodes, CPUs, and GPUs is challenging, but very much needed. Traditional tools simply fail at that scale. This talk focuses on framing the uncertainty analysis and visualization needs against real-world challenges.
The typical approach adopted today to explore simulation uncertainty is to post-process the output from multiple simulation runs and compute the uncertainty. These outputs may not have been obtained together but pieced together from multiple simulation runs. The results computed are often in the form of boxplots, mean-spread plots, RMSE values, and other discipline-specific approaches. Often, workflow tools are employed to manage the execution of the simulation ensemble. However, majority of the data generated is unused and summary approaches miss potentially interesting simulation artifacts. Most of the data sits offline and requires reloading for any analysis.
This talk will discuss the various elements in an end-to-end workflow for understanding and visualizing simulation uncertainty.
About this LectureNumber of Slides: 35
Duration: 50 + 60 minutes
Languages Available: English
Request this Lecture
To request this particular lecture, please complete this online form.
Request a Tour
To request a tour with this speaker, please complete this online form.
All requests will be sent to ACM headquarters for review.