Grant: Discovery Grants Program - Individual

BACKGROUND. Project CellNet (Kharma et al., 2004) is one of a handful of research projects that aim at producing generic Pattern Recognition systems, which can be configured for particular recognition tasks, using Genetic Programming (GP) and some measure of training.   MOTIVATION. The current version of CellNet (CellNet-CoEv) is still an intermediate step on the path of automatic synthesis of a complete multi-classifier. The features are still hard-coded: we intend to replace them with a parameterless GP-based feature creation module. Also, the next version of CellNet will produce complete multi classifiers. METHOD. We intend to maintain the overall emphasis on the use of Genetic Programming for the purpose of feature creation and selection, as well as classifier synthesis. However, we will  introduce other techniques to the next version of CellNet. Some serious pre-processing will be carried out to create a prototypical sum image per class of patterns. This will be followed by the automatic creation via GP, evaluation via statistical techniques, and employment of feature detectors. Finally, the selected features will be used by a classifier, which will be thoroughly assessed using standard pattern databases. CONTRIBUTIONS. The main anticipated contributions are: (a) The production of a fully evolvable multi-class pattern recognition system, which can deal with a range of real-world symbol/signal classification tasks; (b) The development of some leading-edge Genetic Programming techniques, suited to the problem of automatic Feature Creation, in the area of Pattern Recognition; (c) A better understanding of how complexity can arise from simpler forms, given the right mechanisms and initial conditions. REFERENCES. Kharma, N., Kowaliw, T., Clement, E., Jensen, C., Youssef, A., and Yao, J. "Project CellNet: Evolving an Autonomous Pattern Recognizer", International Journal of Pattern Recognition and Artificial Intelligence, Vol. 18, No. 6 (2004) 1-18.