With the ever increasing importance of UQ in various disciplines and fields, software solutions and libraries for UQ problems get more and more important. Progress and use of UQ techniques relies on the availability of software features and support. This raises interesting questions for the UQ community such as: What are the current properties of available tools? For which classes of problems have they been developed? What methods or algorithms do they provide? What are challenges for UQ software and which resources are required? What are recent improvements? What are the next steps and the long-term goals of the development?
This minisymposium brings together experts for different software in the context of UQ, ranging from tools that ease up individual tasks of UQ (such as surrogate modelling, UQ workflows, dimensionality reduction, data augmentation) up to whole frameworks for solving UQ problems. The minisymposium will foster discussion and exchange of ideas between developers and (prospective) users.
16:30
- CANCELED - New Ways to Explore and Predict with Dakota
Brian M. Adams | Sandia National Laboratories | United States
Show details
Authors:
Brian M. Adams | Sandia National Laboratories | United States
J. Adam Stephens | Sandia National Laboratories | United States
Driven by Sandia National Laboratories' applications, the Dakota project (http://dakota.sandia.gov) invests in both state-of-the-art research and robust, usable software for optimization and UQ. Broadly, Dakota's advanced parametric analysis enables sensitivity analysis, design exploration, model calibration, risk analysis, and quantification of margins and uncertainty with computational models. Begun as an internal research project almost 25 years ago, we estimate Dakota now has thousands of users worldwide across academia, government, and industry.
This presentation will emphasize recent Dakota UQ development and research on the horizon. Topics will include Bayesian inference, generalization of multi-level/multi-fidelity UQ with sampling and PCE variants, optimal experimental design, data-driven UQ, as well as Gaussian process and low rank approximations. We will tour user resource improvements that make Dakota UQ more accessible, including screencast videos, examples library, HDF5 output, workflow tools, and the graphical user interface.
17:00
Data-Driven Uncertainty Quantification with SG++
Dirk Pflüger | University of Stuttgart | Germany
Show details
Authors:
Dirk Pflüger | University of Stuttgart | Germany
Michael Rehme | University of Stuttgart | Germany
Adaptive sparse grids provide a flexible and versatile way to
represent higher-dimensional dependencies. We present recent
developments of and with SG++, the most extensive software toolkit for
spatially adaptive sparse grids. It is a multi-platform toolkit with
fast and efficient algorithms. In the context of UQ, it provides in
particular sparse grid functionality for forward propagation, model
calibration, optimization, and density estimation.
Recent extensions include advanced higher-order basis functions, such
as different variants of B-splines, which enable gradient-based
approaches without further approximation, and improved density
estimation. We discuss properties of SG++ and present UQ results for a
CO2 benchmark problem from the Cluster of Excellence ""Data-Integrated
Simulation Science"", for which we have recently studied and compared several UQ approaches.
17:30
Recent Development in the PSUADE UQ Software
Charles H. Tong | Lawrence Livermore National Laboratory | United States
Show details
Author:
Charles H. Tong | Lawrence Livermore National Laboratory | United States
This presentation will describe ongoing development work for PSUADE
(Problem Solving environment for Uncertainty Analysis and Design
Exploration), which is a general-purpose software tool for non-intrusive
('black box' simulation models) uncertainty quantification. Specifically,
After a brief overview of PSUADE's core capabilities, new capabilities
such as optimal experimental design, UQ methods for multi-agent models,
and other dimension reduction methods will be presented.
18:00
Blurring the lines between UQ and ML: a software perspective
Stefano Marelli | ETH Zurich | Switzerland
Show details
Authors:
Stefano Marelli | ETH Zurich | Switzerland
Bruno Sudret | ETH Zurich | Switzerland
Arguably one of the main players of the "data-driven revolution" in the past decades, machine learning (ML) is now ubiquitous in all fields of science and engineering. It is therefore not surprising that many of its core tools, from dimensionality reduction, validation techniques to data-driven modeling, have become part of the uncertainty quantification (UQ) kit. But the increasingly tighter connection between the two does not stop at sharing computational tools: statistics-infused machine learning is now as common as data-driven uncertainty quantification.
In this contribution, we share how our perception of these two formerly disjoint worlds has changed thanks to our experience developing and supporting UQLab, a general-purpose uncertainty quantification platform, in both academic and industrial applications. We also discuss how vastly different fields of science and engineering all share similar needs that can be fulfilled by either ML or UQ, and how we see the software landscape evolving in the upcoming decade.