Top content

Research Interests

I am interested in how to observe and extract knowledge learned by neural networks, thus moving their application from being powerful black boxes for specific tasks to being accessible by logical reasoning. To provide the tools for this challenge, I'm looking for a mathematical foundation which is intuitively accessible at the same time.

Current Project

The mathematical tool of Conceptors (as introduced by H. Jaeger, 2014) promises to extract the dynamics of a reservoir driven by an external stimulus in a way that can not only be used for classification tasks but can also be manipulated with logical operators, thus creating a conceptual representation of the observed stimulus. The project is part of the RTG "Computational Cognition" and aims at a deeper understanding of Conceptor learning and logic as well as applications. It is co-supervised by Prof. Dr. Gordon Pipa and Prof. Dr. Kai-Uwe Kühnberger.