Navigationsweiche Anfang

Navigationsweiche Ende

Select language

Lehrstuhl für Technologien und Management der Digitalen Transformation


Herr Univ.-Prof. Dr.-Ing. Tobias Meisen

Transparent and Interpretable AI

One research focus of the chair TMDT is in the field of Interpretable AI. The driving factor for our research is the need for absolute transparency and traceability of the decision-making processes of an AI within its use for applications in industrial manufacturing.

Our research uses methods from the field of neuroscience to study processes in the brain and aims at the adaptation and transfer of these methods for the investigation of artificial neural networks (ANN). The focus is on the investigation of learned representations within ANNs with respect to their organization and structure. With the help of ablation studies, a procedure inspired from the field of neuroscience for the controlled damage of brain structures, and the transfer of the procedure to ANNs, both close similarities and clear differences in the organization of ANNs to known structures in the brain can be revealed. The understanding of the organization of learned representations in ANNs and the resulting increased degree of transparency for their decision-making processes create confidence in the technology and make AI interpretable for humans. Especially in application areas such as manufacturing or assembly from the field of production engineering, traceability enables the use of AI as assistance systems for the human domain expert.

 

Contact

Richard Meyes, M.Sc.

Interested in this research?

Would you like to write a thesis in this research area? Then look here for open topics or contact us at meyes{at}uni-wuppertal.de.

Would you like to delve deeper into this field? Then join our team as a research assistant! Further information here.

Selected relevant publications

References
Richard Meyes; Constantin Waubert de Puiseau; Andres Posada-Moreno; Tobias Meisen
Under the Hood of Neural Networks: Characterizing Learned Representations by Functional Neuron Populations and Network Ablations
Richard Meyes; Moritz Schneider; Tobias Meisen
How Do You Act? An Empirical Study to Understand Behavior of Deep Reinforcement Learning Agents
Richard Meyes; Melanie Lu; Constantin Waubert de Puiseau; Tobias Meisen
Ablation Studies to Uncover Structure of Learned Representations in Artificial Neural Networks
Proceedings of the 2019 International Conference on Artificial Intelligence (ICAI),
2019
Richard Meyes; Melanie Lu; Constantin Waubert de Puiseau; Tobias Meisen
Ablation Studies in Artificial Neural Networks
arXiv arXiv:1901.08644,
2019
Peter Lillian; Richard Meyes; Tobias Meisen
Ablation of a Robot's Brain: Neural Networks Under a Knife
arXiv arXiv:1812.05687,
2018
Total:
5
Export as:
BibTeX, XML