Pith. sign in

REVIEW 1 cited by

Thompson Sampling in Non-Episodic Restless Bandits

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 1910.05654 v1 pith:7L5MO3JS submitted 2019-10-12 cs.LG stat.ML

Thompson Sampling in Non-Episodic Restless Bandits

classification cs.LG stat.ML
keywords algorithmanalysispolicyrestlessbanditsbounddistributionsmapping
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Restless bandit problems assume time-varying reward distributions of the arms, which adds flexibility to the model but makes the analysis more challenging. We study learning algorithms over the unknown reward distributions and prove a sub-linear, $O(\sqrt{T}\log T)$, regret bound for a variant of Thompson sampling. Our analysis applies in the infinite time horizon setting, resolving the open question raised by Jung and Tewari (2019) whose analysis is limited to the episodic case. We adopt their policy mapping framework, which allows our algorithm to be efficient and simultaneously keeps the regret meaningful. Our algorithm adapts the TSDE algorithm of Ouyang et al. (2017) in a non-trivial manner to account for the special structure of restless bandits. We test our algorithm on a simulated dynamic channel access problem with several policy mappings, and the empirical regrets agree with the theoretical bound regardless of the choice of the policy mapping.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Learning in Markovian bandits with non-observable states and constrained decision epochs

    cs.LG 2026-06 unverdicted novelty 6.0

    Introduces self-degrading Markovian bandits and UCB-NOM algorithm achieving nearly logarithmic regret without prior knowledge and O(log T) with bias bounds, with bounds independent of state count.