Fast UCB-type algorithms for stochastic bandits with heavy and super heavy symmetric noise

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Dorn, Yuriy, Katrutsa, Aleksandr, Latypov, Ilgam, Pudovikov, Andrey
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914674360975360
author Dorn, Yuriy
Katrutsa, Aleksandr
Latypov, Ilgam
Pudovikov, Andrey
author_facet Dorn, Yuriy
Katrutsa, Aleksandr
Latypov, Ilgam
Pudovikov, Andrey
contents In this study, we propose a new method for constructing UCB-type algorithms for stochastic multi-armed bandits based on general convex optimization methods with an inexact oracle. We derive the regret bounds corresponding to the convergence rates of the optimization methods. We propose a new algorithm Clipped-SGD-UCB and show, both theoretically and empirically, that in the case of symmetric noise in the reward, we can achieve an $O(\log T\sqrt{KT\log T})$ regret bound instead of $O\left (T^{\frac{1}{1+α}} K^{\fracα{1+α}} \right)$ for the case when the reward distribution satisfies $\mathbb{E}_{X \in D}[|X|^{1+α}] \leq σ^{1+α}$ ($α\in (0, 1])$, i.e. perform better than it is assumed by the general lower bound for bandits with heavy-tails. Moreover, the same bound holds even when the reward distribution does not have the expectation, that is, when $α<0$.
format Preprint
id arxiv_https___arxiv_org_abs_2402_07062
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Fast UCB-type algorithms for stochastic bandits with heavy and super heavy symmetric noise
Dorn, Yuriy
Katrutsa, Aleksandr
Latypov, Ilgam
Pudovikov, Andrey
Machine Learning
Optimization and Control
In this study, we propose a new method for constructing UCB-type algorithms for stochastic multi-armed bandits based on general convex optimization methods with an inexact oracle. We derive the regret bounds corresponding to the convergence rates of the optimization methods. We propose a new algorithm Clipped-SGD-UCB and show, both theoretically and empirically, that in the case of symmetric noise in the reward, we can achieve an $O(\log T\sqrt{KT\log T})$ regret bound instead of $O\left (T^{\frac{1}{1+α}} K^{\fracα{1+α}} \right)$ for the case when the reward distribution satisfies $\mathbb{E}_{X \in D}[|X|^{1+α}] \leq σ^{1+α}$ ($α\in (0, 1])$, i.e. perform better than it is assumed by the general lower bound for bandits with heavy-tails. Moreover, the same bound holds even when the reward distribution does not have the expectation, that is, when $α<0$.
title Fast UCB-type algorithms for stochastic bandits with heavy and super heavy symmetric noise
topic Machine Learning
Optimization and Control
url https://arxiv.org/abs/2402.07062