Invited Talk: Learning with Sample-Dependent Hypothesis Sets