PAC-BayesAnalysisBeyondtheUsualBounds

We focus on a stochastic learning model where the learner observes a finite set of training examples and the output of the learning process is a data-dependent distribution over a space of hypotheses.