Occam's hammer: a link between randomized learning and multiple testing FDR control

Gilles Blanchard, François Fleuret

We establish a generic theoretical tool to construct probabilistic bounds for algorithms where the output is a subset of objects from an initial pool of candidates (or more general…