Abstract

Current trends in information technology show the necessity to develop much more efficient methods for structural data analysis and data mining. At the same time, these directions open new perspectives for further developments in computer science. There we see the opportunity to capitalize on the experience of the team in the field of neural networks and restarting automata. Several types of problems from the area of structural pattern recognition and data mining are namely of a similar character and the principles of learning from examples using neural networks or restarting automata can be used to approach their solution.

Our goals are

• Design new methods for efficient knowledge extraction. Extend these methods to extract also the information concerning the (hierarchical) structure of the data (using self-organization and sensitivity analysis in BP-networks).
• Extend the original 1D model of restarting automata to work also on 2D-inputs, e.g. pictures. Analyze theoretical properties of 2D-restarting automata, and 2D-grammars, respectively.
• Implement the developed methods and test them with the aim to assess their limits in practical applications. Use the developed software modules in two pilot studies – for knowledge extraction (e.g. from economic or multimedia data) and in structural recognition of mathematical formulae.