Mission

The Information Theory of Cognitive Systems group aims at understanding
learning processes of cognitive systems in terms of information theory.
The research is guided by the belief that these processes are consistent
with information-theoretic optimization principles referred to as infomax
principles within theoretical neuroscience. Several classical approaches
along these lines do provide explanations for experimental findings such
as the development of receptive fields in the primary visual cortex.
However, these approaches usually suffer from an intrinsic restriction on
network and model class primarily addressing only networks with
feed-forward structure similar to the unidirectional sender-receiver
setting of classical information theory. The group is working on
extensions that include various kinds of essential feedbacks. On one hand,
feedbacks are predominant within brain structures such as in
thalamo-cortical interactions including the interaction between LGN and
visual cortex. On the other hand, the sensori-motor integration involves
feedbacks that are of utmost importance for information representation,
information processing, and control. In this sense, the group aims at
combining information-theoretic approaches within theoretical neuroscience
with the field of embodied artificial intelligence, thereby relating
information flows through neuronal systems to system-theoretic concepts.
We feel confident that such a holistic approach addressing feedbacks
at all system levels is necessary for understanding cognition.

Projects

Embodied Artificial Intelligence | This project focusses on different aspects of the sensori-motor loop in conjunction with information theory. The goal is to develop a mathematical understanding of how first principles of learning lead to cognitive capabilities. |

Exponential Families & Information Maximization | In this project the geometry of exponential families is studied from the perspective of information geometry and algebraic statistics. The results are of particular relevance in the context of learning theory. |

Design of Learning Systems |
The goal of this project is to provide mathematical knowledge about various connectionistic models and to identify distinguished architectures of learning systems based on their expressive power and learning performance. |

Robustness of Functional Networks |
The aim of this project is to develop a mathematical formalism of functional robustness and to demonstrate its utility in the context of biological networks. |

Information Theory in Causal Inference |
This project studies how stochastic dependence is generated by causal interactions. It sets constraints on the underlying causal structure in terms of information-theoretic inequality relations. |

Geometry & Complexity | This project develops a geometric understanding of complexity. The aim of this approach is to relate various well-known concepts to each other. |