We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

math.OC

Change to browse by:

References & Citations

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Mathematics > Optimization and Control

Title: A PAC algorithm in relative precision for bandit problem with costly sampling

Abstract: This paper considers the problem of maximizing an expectation function over a finite set, or finite-arm bandit problem. We first propose a naive stochastic bandit algorithm for obtaining a probably approximately correct (PAC) solution to this discrete optimization problem in relative precision, that is a solution which solves the optimization problem up to a relative error smaller than a prescribed tolerance, with high probability. We also propose an adaptive stochastic bandit algorithm which provides a PAC-solution with the same guarantees. The adaptive algorithm outperforms the mean complexity of the naive algorithm in terms of number of generated samples and is particularly well suited for applications with high sampling cost.
Subjects: Optimization and Control (math.OC); Machine Learning (cs.LG); Machine Learning (stat.ML)
DOI: 10.1007/s00186-022-00769-x
Cite as: arXiv:2007.15331 [math.OC]
  (or arXiv:2007.15331v2 [math.OC] for this version)

Submission history

From: Arthur Macherey [view email]
[v1] Thu, 30 Jul 2020 09:22:25 GMT (725kb,D)
[v2] Tue, 12 Apr 2022 12:31:44 GMT (870kb,D)

Link back to: arXiv, form interface, contact.