Learning with Neural Methods Computational Intelligence - Learning with Neural Methods on Structured Data

Research

Our main aim is to develop non-standard neural methods which can process structured data, i.e. high-dimensional and heterogeneous data or data of a priori unlimited size such as sequences, tree-structures, and graphs. The set of neural methods considered in our group so far contains supervised recurrent and recursive networks, prototype based clustering algorithms such as Learning Vector Quantization, self-organizing methods such as (recurrent and recursive) Self-organizing Maps and Neural Gas, and Support Vector Machines. Areas of application are satellite image processing, time series analysis, natural language processing, bioinformatics, and operations research.

Online publications can be found at the homepages of the group members:

Gersmann - Hammer - Strickert

Projects:

  • General recursive self-organizing networks

  • SRNG: prototype based clustering with automatic metric adaptation

  • BB-Trees: rule extraction from prototype based clustering networks

  • SVM for search problems in operations research

  • Theory of feedforward and recurrent/recursive networks

    LNM - Computer Science - University of Osnabrück.

    B.Hammer