Prof. Dr.

Wrobel, Stefan

Filters
Reset filters

Settings
Now showing 1 - 3 of 3
  • Publication
    The why and how of trustworthy AI
    ( 2022-09-03)
    Schmitz, Anna
    ;
    Akila, Maram
    ;
    Hecker, Dirk
    ;
    Poretschkin, Maximilian
    ;
    Wrobel, Stefan
    Artificial intelligence is increasingly penetrating industrial applications as well as areas that affect our daily lives. As a consequence, there is a need for criteria to validate whether the quality of AI applications is sufficient for their intended use. Both in the academic community and societal debate, an agreement has emerged under the term “trustworthiness” as the set of essential quality requirements that should be placed on an AI application. At the same time, the question of how these quality requirements can be operationalized is to a large extent still open. In this paper, we consider trustworthy AI from two perspectives: the product and organizational perspective. For the former, we present an AI-specific risk analysis and outline how verifiable arguments for the trustworthiness of an AI application can be developed. For the second perspective, we explore how an AI management system can be employed to assure the trustworthiness of an organization with respect to its handling of AI. Finally, we argue that in order to achieve AI trustworthiness, coordinated measures from both product and organizational perspectives are required.
  • Publication
    A Novel Regression Loss for Non-Parametric Uncertainty Optimization
    ( 2021)
    Sicking, Joachim
    ;
    Akila, Maram
    ;
    Pintz, Maximilian
    ;
    Wirtz, Tim
    ;
    Fischer, Asja
    ;
    Wrobel, Stefan
    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.