An approach to predicting non-deterministic neural network behavior

This paper describes a methodology for generating indicators of performance for the dynamic cell structures neural network, a type of growing self-organizing map. The performance indicators are based on the learning architecture of the neural network and are validated using correlation measures of Murphy's rule. Time estimates for neural network convergence are generated based on the current data conditions and the confidence in the neural network, which is provided by the performance indicators. Analytical and experimental results are presented for the dynamic cell structures neural network during its training from the Carnegie Mellon University two-spirals benchmark data.