Learning-Augmented Algorithms for MTS with Bandit Access to Multiple Predictors

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Coşa, Matei Gabriel, Eliáš, Marek
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909639633797120
author Coşa, Matei Gabriel
Eliáš, Marek
author_facet Coşa, Matei Gabriel
Eliáš, Marek
contents We consider the following problem: We are given $\ell$ heuristics for Metrical Task Systems (MTS), where each might be tailored to a different type of input instances. While processing an input instance received online, we are allowed to query the action of only one of the heuristics at each time step. Our goal is to achieve performance comparable to the best of the given heuristics. The main difficulty of our setting comes from the fact that the cost paid by a heuristic at time $t$ cannot be estimated unless the same heuristic was also queried at time $t-1$. This is related to Bandit Learning against memory bounded adversaries (Arora et al., 2012). We show how to achieve regret of $O(\text{OPT}^{2/3})$ and prove a tight lower bound based on the construction of Dekel et al. (2013).
format Preprint
id arxiv_https___arxiv_org_abs_2506_05479
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Learning-Augmented Algorithms for MTS with Bandit Access to Multiple Predictors
Coşa, Matei Gabriel
Eliáš, Marek
Machine Learning
Data Structures and Algorithms
68W40 (Primary), 68T05 (Secondary)
We consider the following problem: We are given $\ell$ heuristics for Metrical Task Systems (MTS), where each might be tailored to a different type of input instances. While processing an input instance received online, we are allowed to query the action of only one of the heuristics at each time step. Our goal is to achieve performance comparable to the best of the given heuristics. The main difficulty of our setting comes from the fact that the cost paid by a heuristic at time $t$ cannot be estimated unless the same heuristic was also queried at time $t-1$. This is related to Bandit Learning against memory bounded adversaries (Arora et al., 2012). We show how to achieve regret of $O(\text{OPT}^{2/3})$ and prove a tight lower bound based on the construction of Dekel et al. (2013).
title Learning-Augmented Algorithms for MTS with Bandit Access to Multiple Predictors
topic Machine Learning
Data Structures and Algorithms
68W40 (Primary), 68T05 (Secondary)
url https://arxiv.org/abs/2506.05479