Aaditya Ramdas

Data-adaptive stopping for iterative algorithms is now a general recipe, not a case-by-case puzzle.

Scaling a numeraire e-variable by an independent uniform multiplier guarantees a head-to-head win against any rival test with at least 50% probability, yet fundamentally degrades sequential power and expected log-growth.

Upper-bounding the number of distribution shifts in sequential data is provably impossible without parametric assumptions, but conformal inference can uniquely deliver valid, non-trivial lower bounds.

By cleverly hedging between Cover's and Robbins' betting strategies, you can achieve almost-sure $O(\ln \ln n)$ regret without sacrificing the $O(\ln n)$ worst-case guarantee.