The content on this page was translated automatically.
VDI Lecture: Quality Guarantees and Certification for Machine Learning Methods - Strategies of the Machine Learning Competence Center Rhine-Ruhr
The popularity of Artificial Intelligence (AI) in media and politics leads to a gold rush-like mood when it comes to the use of AI techniques. Although impressive results can indeed be achieved in some application areas, the overestimation of AI methods by users carries uncontrollable risks. Both the resources required and the thoroughness of validation are often not questioned outside of expert circles.
In this talk, strategies of the Machine Learning Competence Center Rhine-Ruhr (ML2R) will be presented, which aim at making application of AI methods certifiable and transparent, thus counteracting the above mentioned problems:
1) Theoretical foundations for AI methods will be established, which allow us to provide guarantees for interpretability, goodness, resource consumption and real-time behavior.
2) The human-centered design of learning methods should achieve understandability, comprehensibility, validability, and even certification.
3) Since in practical applications of machine learning often complex prior knowledge about the application exists, for example in the form of literature, simulations, physical equations or other formal models, it is investigated how such knowledge can be explicitly included in the learning process.
In each case, the strategies will be explained with reference to concrete research results and applications.
Dr. Nico Piatkowski is a senior researcher at the Machine Learning Competence Center Rhine-Ruhr (ML2R) at TU Dortmund University. His research interests include probabilistic machine learning and machine learning under resource constraints.
Free admission.