Very interesting! I'd like to see more information about the structure of the "representatives" you fed the learning program, and some references to the underlying QV algorithm.