scholarly journals On forward and inverse uncertainty quantification for models involving hysteresis operators

2020 ◽  
Vol 15 ◽  
pp. 53
Author(s):  
Olaf Klein ◽  
Daniele Davino ◽  
Ciro Visone
Keyword(s):  
Weight Function ◽  
Uncertainty Quantification ◽  
Real World ◽  
Yield Limit ◽  
Play Operator ◽  
Stochastic Yield ◽  
Hysteresis Operators

Parameters within hysteresis operators modeling real world objects have to be identified from measurements and are therefore subject to corresponding errors. To investigate the influence of these errors, the methods of Uncertainty Quantification (UQ) are applied. Results of forward UQ for a play operator with a stochastic yield limit are presented. Moreover, inverse UQ is performed to identify the parameters in the weight function in a Prandtl-Ishlinskiĭ operator and the uncertainties of these parameters.

scholarly journals Empirical Frequentist Coverage of Deep Learning Uncertainty Quantification Procedures

Entropy ◽  
2021 ◽  
Vol 23 (12) ◽  
pp. 1608
Author(s):  
Benjamin Kompa ◽  
Jasper Snoek ◽  
Andrew L. Beam
Keyword(s):  
Deep Learning ◽  
Uncertainty Quantification ◽  
Real World ◽  
Large Scale ◽  
Prediction Interval ◽  
Brier Score ◽  
Distribution Data ◽  
Calibration Error ◽  
Learning Models ◽  
Frequentist Coverage

Uncertainty quantification for complex deep learning models is increasingly important as these techniques see growing use in high-stakes, real-world settings. Currently, the quality of a model’s uncertainty is evaluated using point-prediction metrics, such as the negative log-likelihood (NLL), expected calibration error (ECE) or the Brier score on held-out data. Marginal coverage of prediction intervals or sets, a well-known concept in the statistical literature, is an intuitive alternative to these metrics but has yet to be systematically studied for many popular uncertainty quantification techniques for deep learning models. With marginal coverage and the complementary notion of the width of a prediction interval, downstream users of deployed machine learning models can better understand uncertainty quantification both on a global dataset level and on a per-sample basis. In this study, we provide the first large-scale evaluation of the empirical frequentist coverage properties of well-known uncertainty quantification techniques on a suite of regression and classification tasks. We find that, in general, some methods do achieve desirable coverage properties on in distribution samples, but that coverage is not maintained on out-of-distribution data. Our results demonstrate the failings of current uncertainty quantification techniques as dataset shift increases and reinforce coverage as an important metric in developing models for real-world applications.

Developing a Numerical Model of the Virginia Tech Stability Wind Tunnel for Uncertainty Quantification Based On Real-World Geometry

2020 ◽  
Author(s):  
Máté Szoke ◽  
Vidya Vishwanathan ◽  
Tim Loeschen ◽  
Aldo Gargiulo ◽  
Daniel J. Fritsch ◽  
...  
Keyword(s):  
Numerical Model ◽  
Wind Tunnel ◽  
Uncertainty Quantification ◽  
Real World ◽  
Virginia Tech

A weighted log-rank test for comparing two survival curves

2020 ◽  
Vol 26 (3) ◽  
pp. 253-262
Author(s):  
Seung-Hwan Lee ◽  
Eun-Joo Lee
Keyword(s):  
Weight Function ◽  
Real World ◽  
Null Hypothesis ◽  
New Method ◽  
Rank Test ◽  
Survival Curves ◽  
Rank Tests ◽  
Asymptotically Normal ◽  
Real World Application ◽  
Log Rank Test

AbstractThis paper proposes a weighted log-rank test that maintains sensitivity to realistic alternatives of two survival curves, such as crossing curves, in the presence of heavy censoring. The new test incorporates a weight function that changes over the censoring level, increasing adaptivity and flexibility of the commonly used weighted log-rank tests. The new statistic is asymptotically normal under the null hypothesis that there is no difference in survival between two groups. The performances of the new test are evaluated via simulations under both proportional and non-proportional alternatives. We illustrate the new method with a real-world application.

Progress Toward a Complete Set of Errors for Modeling and Simulation

2020 ◽  
Vol 5 (3) ◽  
Author(s):  
Joshua Kaizer
Keyword(s):  
Modeling And Simulation ◽  
Uncertainty Quantification ◽  
Real World ◽  
Total Error ◽  
World System ◽  
Distinct Difference ◽  
Correct Prediction ◽  
Step Process ◽  
The Real ◽  
Complete Set

Abstract To develop a fully complete set of errors associated with modeling and simulation, it is necessary to express every error that could impact the accuracy of a computational model's prediction of the real world system (i.e., a set of errors that is theoretically complete) and to develop a means to assess each error (i.e., making the set practically complete). As a first step toward this goal, this paper focuses on developing a theoretically complete set of errors that, if accounted for, would result in the correct prediction of reality. In order to derive this theoretically complete set of errors, a three-step process is followed. First, a generic scenario is introduced which is defined by a set of functions and inputs common to many, if not most, applications in modeling and simulation. Second, using only these functions and inputs, an equation for the total error is defined such that correcting the model's prediction to account for the error would result in a correct prediction of reality. Finally, the equation for total error is expanded by introducing terms from the generic scenario. This results in a decomposition of the total error into a set of thirteen distinct difference terms, each of which is defined as an error and many of which are closely related to current practices in verification, validation, and uncertainty quantification. These thirteen errors represent a theoretically complete set.

Multiscale Validation and Uncertainty Quantification for Problems With Sparse Data

2017 ◽  
Vol 2 (1) ◽  
Author(s):  
Anchal Jatale ◽  
Philip J. Smith ◽  
Jeremy N. Thornock ◽  
Sean T. Smith ◽  
Jennifer P. Spinti ◽  
...  
Keyword(s):  
Experimental Data ◽  
Wind Tunnel ◽  
Uncertainty Quantification ◽  
Real World ◽  
Ghg Emissions ◽  
Sparse Data ◽  
World Systems ◽  
Global Approach ◽  
Real World Problem ◽  
Helium Plume

Quantification of uncertainty in the simulation results becomes difficult for complex real-world systems with little or no experimental data. This paper describes a validation and uncertainty quantification (VUQ) approach that integrates computational and experimental data through a range of experimental scales and a hierarchy of complexity levels. This global approach links dissimilar experimental datasets at different scales, in a hierarchy, to reduce quantified error bars on case with sparse data, without running additional experiments. This approach was demonstrated by applying on a real-world problem, greenhouse gas (GHG) emissions from wind tunnel flares. The two-tier validation hierarchy links, a buoyancy-driven helium plume and a wind tunnel flare, to increase the confidence in the estimation of GHG emissions from wind tunnel flares from simulations.

Replications can cause distorted belief in scientific progress

2018 ◽  
Vol 41 ◽  
Author(s):  
Michał Białek
Keyword(s):  
Real World ◽  
Scientific Progress ◽  
The Public ◽  
Psychological Science

AbstractIf we want psychological science to have a meaningful real-world impact, it has to be trusted by the public. Scientific progress is noisy; accordingly, replications sometimes fail even for true findings. We need to communicate the acceptability of uncertainty to the public and our peers, to prevent psychology from being perceived as having nothing to say about reality.

Current Issues: Evidence-Based Practice for the Real World: Thoughts on Clinical Education in Communication Disorders

2010 ◽  
Vol 20 (3) ◽  
pp. 100-105 ◽  
Author(s):  
Anne K. Bothe
Keyword(s):  
Clinical Education ◽  
Real World ◽  
Evidence Based Practice ◽  
Communication Disorders ◽  
University Faculty ◽  
Evidence Based ◽  
Speech Language Pathology ◽  
Basic Principles ◽  
The Real ◽  
Language Pathology

This article presents some streamlined and intentionally oversimplified ideas about educating future communication disorders professionals to use some of the most basic principles of evidence-based practice. Working from a popular five-step approach, modifications are suggested that may make the ideas more accessible, and therefore more useful, for university faculty, other supervisors, and future professionals in speech-language pathology, audiology, and related fields.

Sign in / Sign up

Export Citation Format

Share Document