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


Current browse context:


Change to browse by:

References & Citations


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

Mathematics > Statistics Theory

Title: On Explore-Then-Commit Strategies

Authors: Aurélien Garivier (IMT), Emilie Kaufmann (SEQUEL, CRIStAL, CNRS), Tor Lattimore
Abstract: We study the problem of minimising regret in two-armed bandit problems with Gaussian rewards. Our objective is to use this simple setting to illustrate that strategies based on an exploration phase (up to a stopping time) followed by exploitation are necessarily suboptimal. The results hold regardless of whether or not the difference in means between the two arms is known. Besides the main message, we also refine existing deviation inequalities, which allow us to design fully sequential strategies with finite-time regret guarantees that are (a) asymptotically optimal as the horizon grows and (b) order-optimal in the minimax sense. Furthermore we provide empirical evidence that the theory also holds in practice and discuss extensions to non-gaussian and multiple-armed case.
Subjects: Statistics Theory (math.ST); Machine Learning (cs.LG)
Cite as: arXiv:1605.08988 [math.ST]
  (or arXiv:1605.08988v2 [math.ST] for this version)

Submission history

From: Emilie Kaufmann [view email]
[v1] Sun, 29 May 2016 10:35:33 GMT (641kb,D)
[v2] Mon, 14 Nov 2016 12:40:20 GMT (322kb,D)

Link back to: arXiv, form interface, contact.