Brain Computer Interfaces
We use supervised machine learning techniques to detect event-related potentials (ERPs) in a human's electroencephalogram (EEG) that indicate that the human has perceived and recognized important messages that have been presented to him or that he intends to execute a movement shortly. My work in the VI-Bot and IMMI project was focused mainly on investigating how the system can detect these ERPs with a minimum amount of labeled training data from the current user by reusing data from historic sessions of the same and other users. [Relevant publications]
Neuroevolution
We have developed a new neuroevolutionary method called Evolutionary Acquisition of Neural Topologies (EANT) and a new genetic encoding, the common genetic encoding (CGE). We have proved certain theoretical properties of the encoding and have conducted empirical evaluations of EANT in different Reinforcement Learning problems. [Relevant publications]
Graph Matching
We have used methods from structural pattern recognition to develop means to automatically determine suitable landmarks in three-dimensional images (CT, MRT) of liver and lung that allow to registrate images of the same organ taken at different points in time or with different modalities. For this, we have developed new approaches for matching vessel trees extracted from the images and used the associated tree branching points as landmarks. This work was mainly done as part of my diploma thesis at the "Computer Vision and Pattern Recognition Group" of the University of Muenster and in cooperation with MeVis Research (now Fraunhofer MEVIS). [Relevant publications]
MMLF: The Maja Machine Learning Framework (MMLF) is a general framework for problems in the domain of Reinforcement Learning (RL) written in python. It provides a set of RL related algorithms and a set of benchmark domains. Furthermore it is easily extensible and allows to automate benchmarking of different agents. Among the RL algorithms are TD(lambda), CMA-ES, Fitted R-Max, Monte-Carlo learning, the DYNA-TD and the actor-critic architecture. MMLF contains different variants of the maze-world and pole-balancing problem class as well as the mountain-car testbed and the pinball maze domain.