5 problems

Let a distribution-free prediction method be applied to a training dataset, and consider its training-conditional coverage guarantee. Bian et al.'s stability conjecture. An assumpt…

Let Nesterov's accelerated gradient method be applied to a general convex and smooth objective, and let uniform stability measure the sensitivity of its output to changes in one op…

Let the counterexample constructed in the paper establish failure of initial-value stability for standard deterministic accelerated gradient descent, even for strongly convex and s…

Nesterov–heavy-ball stability conjecture. The stability bounds of NAG and HB should continue to hold in the general convex smooth setting.

Nesterov stability conjecture. The same stability bound should hold for general convex smooth loss functions.