Research

My research interest is in machine learning and intelligent & complex systems. Here is an overview:

[Neurons]Neural Networks & Machine Learning
At present, my main research focus is on recurrent neural networks (RNN) for prediction and classification. RNNs are particularly well suited for sequential and time-series data. I am investigating efficient training methods, ways to distribute computation in large neural networks, and applications of RNNs, e.g., for robotics or sensor networks. We are also looking at implementations of neural networks on alternative hardware. See my Neural Network page for details....

phase transition Intelligent & Complex Systems
How can we understand and create artificial systems that are self-organised, but still achieve defined goals? What are the underlying mechanisms that lead to specific behaviours? Examples of our work in this area are mechanisms that steer teams of agents towards a goal without explicit communication. This has applications in, e.g., simulating crowds of people, or analysing sport teams. Communication, on the other hand, and how it emerges is also one of our research topics. A related question that we are interested in is how information is processed in cognitive systems. More Intelligent & Complex Systems ...

roboterfussball_100x100 Robot Soccer Creating teams of robots that cooperate to achieve a common goal – or score goals, in the case of robotic soccer – is challenging. RoboCup is a research platform to investigate approaches that make robots cooperate successfully. RoboCup is also a scientific competition and conference, where each year the best of these approaches compete against each other, and new ideas are exchanged. In the past 15 years, we have participated in RoboCup with our teams RoboLog (1999-2006) and Gliders (2012 and 2013). We have also established the 3D soccer simulation league (held since 2004), and the open source simulator SimSpark that is still in use and maintained by the RoboCup community. More...