• English
  • Deutsch
  • Log In
    Password Login
    Research Outputs
    Fundings & Projects
    Researchers
    Institutes
    Statistics
Repository logo
Fraunhofer-Gesellschaft
  1. Home
  2. Fraunhofer-Gesellschaft
  3. Konferenzschrift
  4. A Novel Regression Loss for Non-Parametric Uncertainty Optimization
 
  • Details
  • Full
Options
2021
Conference Paper
Title

A Novel Regression Loss for Non-Parametric Uncertainty Optimization

Abstract
Quantification of uncertainty is one of the most promising approaches to establish safe machine learning. Despite its importance, it is far from being generally solved, especially for neural networks. One of the most commonly used approaches so far is Monte Carlo dropout, which is computationally cheap and easy to apply in practice. However, it can underestimate the uncertainty. We propose a new objective, referred to as second-moment loss (SML), to address this issue. While the full network is encouraged to model the mean, the dropout networks are explicitly used to optimize the model variance. We intensively study the performance of the new objective on various UCI regression datasets. Comparing to the state-of-the-art of deep ensembles, SML leads to comparable prediction accuracies and uncertainty estimates while only requiring a single model. Under distribution shift, we observe moderate improvements. As a side result, we introduce an intuitive Wasserstein distance-based uncertainty measure that is non-saturating and thus allows to resolve quality differences between any two uncertainty estimates.
Author(s)
Sicking, Joachim
Fraunhofer-Institut für Intelligente Analyse- und Informationssysteme IAIS  
Akila, Maram  
Fraunhofer-Institut für Intelligente Analyse- und Informationssysteme IAIS  
Pintz, Maximilian
Universität Bochum
Wirtz, Tim  
Fraunhofer-Institut für Intelligente Analyse- und Informationssysteme IAIS  
Fischer, Asja
Fraunhofer-Institut für Intelligente Analyse- und Informationssysteme IAIS  
Wrobel, Stefan  
Fraunhofer-Institut für Intelligente Analyse- und Informationssysteme IAIS  
Mainwork
3rd Symposium on Advances in Approximate Bayesian Inference, AABI 2021. Online resource  
Project(s)
ML2R
Funder
Bundesministerium für Bildung und Forschung BMBF (Deutschland)  
Bundesministerium für Wirtschaft und Energie BMWi (Deutschland)  
Conference
Symposium on Advances in Approximate Bayesian Inference (AABI) 2021  
File(s)
Download (3.64 MB)
Rights
Use according to copyright law
DOI
10.24406/publica-fhg-411110
Language
English
Fraunhofer-Institut für Intelligente Analyse- und Informationssysteme IAIS  
  • Cookie settings
  • Imprint
  • Privacy policy
  • Api
  • Contact
© 2024