Classifier Technology and the Illusion of Progress

Link to the paper by David J. Hand

Do we need to go deeper? Are we deep enough. Model flexibility, and ability to extract features brought with the advancements within deep learning community are great. However, as these deep models move from "simple" classifciation of cats vs dogs into real-life settings we ought to, at least, acknowledge sources of ucertainty that are not considered in the classical supervised classification paradigm.