Efficient random was my hunch that there should exist efficient random pattern generation. It should not have to repeat usual patterns so often. Maybe the patterns should be different in recognition classification. Like the patterns we use all look noisy. Perhaps we should have a list of different pattern types to test.