Project

ExplAINN

Explainable AI and Neural Networks

Explainable AI and Neural Networks

Despite astonishing progress in the field of Machine Learning (ML), the robustness of high-performance models, especially the ones based on Deep Learning technologies, has been lower than initially predicted. These networks do not generalize as expected, remaining vulnerable to small adversarial perturbations (also known as adversarial attacks). Such shortcomings pose a critical obstacle to implement Deep Learning models for safety-critical scenarios such as autonomous driving, medical imaging, and credit rating.

Moreover, the gap between good performance and robustness also demonstrates the severe lack of explainability for modern AI approaches: Despite good performance, even experts cannot reliably explain model predictions.

Hence, the goals of this project are threefold:

  • Investigate methods of explainability and interpretability for existing AI approaches (focusing on Deep Neural Networks).
  • Develop novel architectures and training schemes that are more interpretable by design.
  • Analyze the trade-offs between explainability, robustness, and performance.

Sponsors

Federal Ministry of Education and Research (BMBF)

01IS19074

Federal Ministry of Education and Research (BMBF)

Share project:

Contact Person
Sebastian Palacio, M.Sc.

Publications about the project

Tobias Hinz, Stanislav Frolov, Federico Raue, Jörn Hees, Andreas Dengel

In: Neural Networks 144 Pages 187-209 Journal of Neural Network Elsevier 12/2021.

To the publication
Mohsin Munir, Sheraz Ahmed, Sebastian Palacio, Adriano Lucieri, Jörn Hees, Andreas Dengel

In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops. International Conference on Computer Vision (ICCV-2021) Online Computer Vision Foundation 10/2021.

To the publication
Philipp Engler, Sebastian Palacio, Jörn Hees, Andreas Dengel

In: International Conference on Pattern Recognition (ICPR) (ICPR) 2020 25th International Conference on Pattern Recognition (ICPR). International Conference on Pattern Recognition (ICPR-2020) Online Pages 8937-8944 1 1 IEEE 5/2021.

To the publication

German Research Center for Artificial Intelligence
Deutsches Forschungszentrum für Künstliche Intelligenz