In this talk, we will describe a few connections between deterministic inequalities that have an "optimization flavor" and probabilistic inequalities for martingales or i.i.d. processes. We will discuss the Burkholder special functions and their role in attaining optimal Rademacher-based regret bounds in online learning.