Agnostic Selective Classification

For a learning problem whose associated excess loss class is (β, B)-Bernstein, we show that it is theoretically possible to track the same classification performance of the best (unknown) hypothesis in our class, provided that we are free to abstain from prediction in some region of our choice.