Nonstochastic Bandits with Composite Anonymous Feedback - Archive ouverte HAL Access content directly
Conference Papers Year : 2018

Nonstochastic Bandits with Composite Anonymous Feedback

(1) , (2, 3) , (4, 3)
1
2
3
4

Abstract

We investigate a nonstochastic bandit setting in which the loss of an action is not immediately charged to the player, but rather spread over at most d consecutive steps in an adversarial way. This implies that the instantaneous loss observed by the player at the end of each round is a sum of as many as d loss components of previously played actions. Hence, unlike the standard bandit setting with delayed feedback, here the player cannot observe the individual delayed losses, but only their sum. Our main contribution is a general reduction transforming a standard bandit algorithm into one that can operate in this harder setting. We also show how the regret of the transformed algorithm can be bounded in terms of the regret of the original algorithm. Our reduction cannot be improved in general: we prove a lower bound on the regret of any bandit algorithm in this setting that matches (up to log factors) the upper bound obtained via our reduction. Finally, we show how our reduction can be extended to more complex bandit settings, such as combinatorial linear bandits and online bandit convex optimization.
Fichier principal
Vignette du fichier
colt2018.pdf (404.69 Ko) Télécharger le fichier
Origin : Files produced by the author(s)
Loading...

Dates and versions

hal-01916981 , version 1 (09-11-2018)

Identifiers

  • HAL Id : hal-01916981 , version 1

Cite

Nicolo Cesa-Bianchi, Claudio Gentile, Yishay Mansour. Nonstochastic Bandits with Composite Anonymous Feedback. COLT 2018 - 31st Annual Conference on Learning Theory, Jul 2018, Stockholm, Sweden. pp.1 - 23. ⟨hal-01916981⟩
65 View
130 Download

Share

Gmail Facebook Twitter LinkedIn More