A Appendix

This proves that vc(C) 1. Analysis By Markov's inequality, it follows that Pr B to have robust loss zero on S, i.e. On each round t T, B is allowed to: 1. We assume w.l.o.g. that the online learner