Wasserstein Distributionally Robust Regret Optimization
The episode discusses the paper "Wasserstein Distributionally Robust Regret Optimization" by Fiechtner and Blanchet from Stanford University. The hosts explain how this approach minimizes regret instead of worst-case loss, showing it can be more aggressive in certain scenarios. They conclude that while empirical risk minimization is optimal in regular cases, regret optimization is useful for nonsmooth losses and larger uncertainty radii, providing both theoretical insights and computational tools.