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.
14:00
Quasi-Optimal Sparse Grids Method for Periodic Functions
Miroslav Stoyanov | Oak Ridge National Laboratory | United States
Show details
Authors:
Miroslav Stoyanov | Oak Ridge National Laboratory | United States
Zachary Morrow | North Carolina State University | United States
We consider adaptive sparse grid interpolation applied to periodic functions of finite differentiability, with application to molecular potential energy surfaces models. Approximation uses a basis of trigonometric polynomials with coefficients computed by multidimensional fast-Fourier-transform. The adaptive procedure infers the decay of the Fourier coefficients using an anisotropic quasi-optimal estimate for the best approximation space. The procedure is implemented in the Tasmanian UQ library, adaptivity is performed using distributed (MPI) asynchronous sampling algorithm.
14:30
Using Chaospy to address stochastically dependent Polynomial Chaos Expansions
Jonathan Feinberg | Expert Analytics | Norway
Show details
Author:
Jonathan Feinberg | Expert Analytics | Norway
Polynomial chaos expansion is a popular method for extracting statistical
metrics from solutions of forward problems with well characterized variability
in the model parameters. Chaospy is a numerical implementation specifically
developed to construct, manipulate and analyze polynomial chaos expansions
using the high modularity and expressiveness of Python. The toolbox allows for
implementation of an expansion using only a handful lines of code.
Chaospy is a development foundry that allows for easy experimentation with
custom features beyond the scope of standard implementation. To demonstrate the
capabilities of the Chaospy toolbox, this talk will demonstrate how polynomial
chaos expansions can be applied on an stochastically dependent random variable.
This includes constructing dependent random variables, and constructing the
uncertainty quantification using a couple of approaches.
15:00
UQEF: A software framework for UQ with automatic, scalable scheduling using Chaospy
Florian Künzner | Technical University of Munich | Germany
Show details
Authors:
Florian Künzner | Technical University of Munich | Germany
Tobias Neckel | Technical University of Munich | Germany
Hans-Joachim Bungartz | Technical University of Munich | Germany
UQ and its variety of methods is a broad field. Users, who want to analyse their applications with respect to (w.r.t.) UQ often perfom rapid prototyping on their local development computers. They need to find a suitable UQ method and appropriate corresponding parameters. Later in the development process, the production runs are typically executed on a cluster, due to the large number of runs that are required. In this talk, we present UQEF (uncertainty quantification execution framework) which is a software framework using Chaospy. UQEF supports rapid prototyping by a proper software design for easy integration of new applications and for automatic scaling from development to production environments via different (optimised) scheduling strategies.
Download presentation
15:30
Uncertainpy: A Python toolbox for uncertainty quantification and sensitivity analysis tailored towards computational neuroscience models
Simen Tennøe | Expert Analytics | Norway
Show details
Authors:
Simen Tennøe | Expert Analytics | Norway
Gaute Einevoll | Norwegian University of Life Sciences and University of Oslo | Norway
Geir Halnes | Norwegian University of Life Sciences | Norway
Computational models in neuroscience typically contain many parameters that either are poorly constrained by experimental data, have an inherent biological variability, or a combination of both. Unfortunately, the application of uncertainty quantification and sensitivity analysis is not yet standard within the field of neuroscience. To help alleviate this problem we have created Uncertainpy, an open-source Python toolbox, tailored to perform uncertainty quantification and sensitivity analysis of neuroscience models.
Uncertainpy aims to make it easy to perform uncertainty quantification and sensitivity analysis without requiring detailed prior knowledge of uncertainty analysis. Uncertainpy uses both polynomial chaos expansions and quasi-Monte Carlo methods from the Chaospy Python package.
Uncertainpy is tailored for neuroscience applications by its built-in capability for calculating characteristic features in the model output. A common feature of computational models in neuroscience is that they contain sharp spikes in the model output. The timing of these spikes is very sensitive to changes in the model parameters, which complicates the uncertainty analysis. Uncertainpy does not only perform a point-to-point comparison of the “raw” model output, but also calculates the uncertainty and sensitivity of salient model response features (e.g spike timing).
This talk will give an overview of Uncertainpy and how it is tailored towards neuroscience.