Multistage Neural Networks

The problem of pattern recognition is one of key issues in machine learning field. Various methods were proposed in literature depending on issues of the problem. In this work the concept of multistage neural networks is going to be presented. This approach is extension of multistage identification. The possibility of using this type of structure for pattern recognition would be discussed and examined with chosen problem from field area. Two-stage writer recognition is going to be considered as practical example in the work. Experiments using well known repository dataset are going to be made to evaluate proposed two-stage approach. On second stage iconic gesture is going to be recognized and on first stage writer is going to be identify in way determined by second stage output. The results of experiment would be confront with other possible methods used for the problem. Possible further directions of the work would be described at the end.

Autorentext

Maciej Zieba was born in Wroc aw in 1985. In 2009 he accomplished MSc Programme in Computer Science at Wroclaw University of Technology (Poland) and Blekinge Institute of Technology (Sweden). Currently he is PhD student at WTU. His scientific interests focus on ensemble methods, pattern recognition and neural networks.