After the completion of my post-doc research position (as project leader of our group here in Würzburg) on the European project "Mind RACES, from reactive to anticipatory cognitive embodied systems", I am now an assistant professor in the framework of the Emmy Noether Program, leading a small research group. We are developing cognitive bodyspaces, that is, spatial representations of the body (or rather simulated body-like structures) within its environment, investigating methods of learning sensorimotor, population-encoded representations for the effective generation of goal-directed behavior. Please visit the lab webpage for further details: COBOSLAB.

Teaching

I am co-organizing a seminar on "Aktuelle Trends in der Künstlichen Intelligenz und Kognitionswissenschaft" (Hot topics in artificial intelligence and cognitive science) and am also co-teaching the Praktikum (laboratory work) on applied artificial intelligence, that is, "Modellierung intelligenter Systeme".
After the successful completion of my class on "Kognitive Systeme" (Cognitive Systems) last summer term, this summer term (2009) I will be teaching Machine Learning of Behavior Control (Maschinelles Lernen von Verhaltensstrukturen). While Cognitive Systems gave an overview on the recent advances and insights in biological systems and the corresponding current modeling efforts in artificial cognitive systems, Machine Learning of Behavior Control will fully focus on how behavior can be learned to be controlled and directed effectively. Again, the lecture is interdisciplinary combining knowledge from psychology and neuroscience with actual modeling approaches and mechanisms from machine learning.

Further Research Interests

Major research interest lies in the analysis and development of anticipatory cognitive systems, that is, systems that self-develop suitable sensorimotor structures in order to efficiently act goal-directedly. Recent research insights, spanning from cognitive psychology, neuroscience, and linguistics to artificial intelligence, suggest that anticipations can be found in a large variety of cognitive mechanisms. It appears that anticipations (in a broad sense) form the basis of effective adaptive behavior as well as (life-long) learning in general. Moreover, I recently postulated that an anticipatory drive underlies, directs, and shapes our emerging inner realities and self perceptions, essentially forming the structural foundations for self-consciousness.
The workshop series on Anticipatory Behavior in Adaptive Learning Systems (ABiALS) is designed to investigate anticipatory mechanisms and structures in detail, analyzing different types of predictive and anticipatory systems and propagating their development. The accompanying books (2003 and 2007) to the workshop series provide comprehensive overviews, including philosophical considerations, conceptualizations, as well as concrete system implementations and evaluations. Complementing the study of anticipatory behavior, I am now focusing also on the design and analysis of self-developing systems. These systems will develop pro-active spatial representations, that is, bodyspaces will develop that are maximally suitable to act efficiently and goal-directedly.

During my PhD time, I had the opportunity to be a research assistant at the Illinois Genetic Algorithms Laboratory (IlliGAL) working together with David E. Goldberg on Genetic Algorithms (GAs) and learning classifier systems. My PhD thesis focuses on online learning with rule-based evolutionary algorithms that combine gradient-based algorithms, such as reinforcement learning techniques, with evolutionary algorithms. While the gradient algorithm is designed to form maximally accurate local predictions, the evolutionary algorithm is designed to distribute these partially overlapping local predictions maximally effectively over the problem space. In combination, the algorithm evolves complete, compact, maximally accurate problem solutions for a variety of problems including function approximation problems, reinforcement learning problems, and classification problems. My book on "Rule-based Evolutionary Online Learning Systems: A Principal Approach to LCS Analysis and Design" investigates these interactive mechanisms in the XCS classifier system using a modular, facet-wise analysis and design approach. XCS is shown to be a highly interactive learning system that can solve a large variety of typical cognitive and adaptive learning problems. Interestingly, the current function-approximation system XCSF develops generalized spatial coverages that are maximally suitable to form accurate predictions. Future research will show in which way XCSF may also be used to form cognitive bodyspaces.

Workshops

I have previously co-organized the International Workshop on Learning Classifier Systems (IWLCS 2007 and IWLCS 2008),