
FTPL with Fréchet perturbations achieves optimal regret in adversarial bandits and Best-of-Both-Worlds in stochastic bandits.


coremsg

ftpl-with-fréchet-type-tail-distributions-optimality-in-adversarial-bandits-and-best-of-both-worlds


FTPL with Fréchet-type Tail Distributions: Optimality in Adversarial Bandits and Best-of-Both-Worlds



Fréchet-type tail distributions enable FTPL to achieve optimal regrets in adversarial bandits and BOBW capability in stochastic bandits.
