BT Group


 Detlef D. Nauck



Memberships in Committees and Societies

My research interests are in the area of fuzzy systems and neural networks. I have developed the NEFCON, NEFCLASS and NEFPROX models which are all available as software tools.

I am working in the Computational Intelligence Group of BT's Intelligent Systems Research Centre. In my current position as a Chief Research Scientist and Technical Group Leader I am involved in a number of projects in the area of Intelligent Data Analysis.

Neuro-Fuzzy Systems

Neuro-fuzzy systems are a way to induce fuzzy systems from data. They are able to learn both fuzzy rules and fuzzy sets. Neuro-fuzzy systems obtained their name, because they apply heuristical learning algorithms that are inspired by artificial neural networks. The learning algorithms operate on local information and cause local modifications, i.e. they do not optimise a fuzzy system globally.

Because interpretability is a key feature of fuzzy systems the learning algorithm of a neuro-fuzzy approach must be constrained such that the semantics of the trained fuzzy system is not affected.

Neuro-fuzzy systems are especially useful for smaller problems where no complex relationships between variables need be modelled. They can use prior knowledge to initialise the learning process but can also learn fuzzy systems from scratch.

My current research in this area focusses on the following issues:

  • handling non-numerical variables,
  • pruning algorithms to simplify solutions,
  • fusing expert knowledge and knowledge induced from data,
  • assessing and guaranteeing the interpretability of a learning outcome.

Intelligent Data Analysis

Data analysis can be described as the process of computing summaries and derived values from data, or - more general - as the process of converting data into knowledge or information. Intelligent Data Analysis (IDA) is a process of critical assessment, exploration, testing and evaluation. It requires the application of knowledge and expertise about the data and it is fundamentally interdisciplinary. In data analysis the emphasis is not on modeling but on answering questions. I am working on identifying suitable analysis techniques from soft computing and machine learning depending on the specific problem and the demands of the user. The goal is to automatically derive interpretable (meaningful) and applicable (useful) models that can be readily put to use in industrial solutions.