Below you find a list of my current research interests. For more information, parallel interests or possible collaboration please do not hesitate to contact me! "A Learning Rule for Universal Approximators with a Single Non-Linearity": Here we develop and investigate learning algorithms for soft-WTA gates. The use and the computational power of WTA-gates are of importance in the area of neural computation and neuromorphic engineering, e.g. for attention in visual systems, but learning algorithms for these gates are largely unknown. "Development of software for real-time spike-based computation on Linux": The micro-robot Khepera has been widely used for experiments for navigation, learning and sensory information processing. Neuromorphic engineering was used to study biological plausible computations with spike trains through devices (e.g. silicon retina) mounted onto the Khepera and Koala. Limitations of possible computations through technical feasibility have been observed. Little effort has been made so far using real-world devices in combination with spike-based computation on a software level of real-time simulations. The goal of this project is to bypass this limitation through providing a Linux based platform for real-time spike-based computations with real-world devices. "Speeding up reinforcement learning for use on real robots": Many attempts have been made to speed-up reinforcement learning. Some of these use the structuring of the state-action space through hierarchies of smaller RL-problems (Dayan&Hinton, Parr&Russel, Kalmar et al). On the other hand there are approaches in robot programming that do not learn at all but provide good ideas of a modularisation of the problem (e.g. Brooks' subsumption architecture). The promising combination of giving a biologically motivated structure to a RL-task in a real-world problem is the goal of this project. Again, a Khepera with a mounted video camera is used as the experimental platform. "Spike-based RL": Recent results from neurobiology suggest that a combination of prediction and reward maybe used in biological system to select actions and to control behaviour (Schulz et al, Read Montague et al.). I am curious about a combination of RL-tasks and spike-based computation. An evolving project could be based on the two projects above and would extend the possibilities of learning and spike-based computation on real-world devices.