After arriving at USM I finally was able to find time to start looking into a subject that I was interested in for a while; Neural Networks (please follow the link to the IJCR journal for the information on the Special Issue devoted to NN's that I am co-editing). Since this is my beginning in this area I am interested in gaining a general feeling for the field by studying the performance characteristics of various NN architectures applied to the simplified character recognition problem. I am particularly interested in applying NNs directly (without pre-processing and/or feature extraction). The preliminary results have been summarized in:

+ "Applying Probabilistic Neural Networks to the Multifont Recognition Problem with Large Training Set," with S. Bowers and A. Costeines, in: A.N. Kumar et. al. (eds.), Proceedings of the Twelfth International Florida AI Research Society Conference, AAAI Press, Menlo Park, CA, 1999, 336-339

+ "Evaluating the Performance of Three Neural Network Architectures Applied to the Pattern Recognition Problem," with S. Bowers, A. Costeines, Proceedings of the 15th Annual Conference on Applied Mathematics, University of Central Oklahoma, Edmond, OK, 1999, 29-36

+ S. Bowers, J. Morrison, M. Paprzycki, Comparing Performance of Neural Networks Recognizing Machine Generated Characters. In: Proceedings of the First Southern Symposium on Computing, Hattiesburg, MS, December, 1998, CD, 12 pages

+ "Applying PNN's to the Multifont Pattern Recognition Problem," submitted to: Proceedings of the 7th Conference on Numerical Methods and Applied Computer Science, Plovdiv, Bulgaria, August, 1998

Copies of the papers can be obtained by e-mailing me at marcin.paprzycki@ibspan.waw.pl