What are the different ways to adress the uncertainty of results provided by Machine Learning models (regression, classification...). Error propagation from the original data? Uncertainty simulation by bootstrap or other methods? Any relevant sources?