A Survey on Data-Dependent Worst-Case Generalization Bounds

Survey organizing recent work on data-dependent worst-case generalization bounds that recover non-vacuous guarantees by restricting attention to parameter space visited by learning algorithms, extending PAC-Bayesian theory.