{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,6,11]],"date-time":"2026-06-11T16:44:09Z","timestamp":1781196249760,"version":"3.54.1"},"reference-count":30,"publisher":"Institute for Operations Research and the Management Sciences (INFORMS)","issue":"3","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Management Science"],"published-print":{"date-parts":[[2021,3]]},"abstract":"<jats:p> The contextual bandit literature has traditionally focused on algorithms that address the exploration\u2013exploitation tradeoff. In particular, greedy algorithms that exploit current estimates without any exploration may be suboptimal in general. However, exploration-free greedy algorithms are desirable in practical settings where exploration may be costly or unethical (e.g., clinical trials). Surprisingly, we find that a simple greedy algorithm can be rate optimal (achieves asymptotically optimal regret) if there is sufficient randomness in the observed contexts (covariates). We prove that this is always the case for a two-armed bandit under a general class of context distributions that satisfy a condition we term covariate diversity. Furthermore, even absent this condition, we show that a greedy algorithm can be rate optimal with positive probability. Thus, standard bandit algorithms may unnecessarily explore. Motivated by these results, we introduce Greedy-First, a new algorithm that uses only observed contexts and rewards to determine whether to follow a greedy algorithm or to explore. We prove that this algorithm is rate optimal without any additional assumptions on the context distribution or the number of arms. Extensive simulations demonstrate that Greedy-First successfully reduces exploration and outperforms existing (exploration-based) contextual bandit algorithms such as Thompson sampling or upper confidence bound. <\/jats:p><jats:p> This paper was accepted by J. George Shanthikumar, big data analytics. <\/jats:p>","DOI":"10.1287\/mnsc.2020.3605","type":"journal-article","created":{"date-parts":[[2020,7,27]],"date-time":"2020-07-27T13:48:37Z","timestamp":1595857717000},"page":"1329-1349","source":"Crossref","is-referenced-by-count":76,"title":["Mostly Exploration-Free Algorithms for Contextual Bandits"],"prefix":"10.1287","volume":"67","author":[{"ORCID":"https:\/\/2.zoppoz.workers.dev:443\/https\/orcid.org\/0000-0002-8793-4732","authenticated-orcid":false,"given":"Hamsa","family":"Bastani","sequence":"first","affiliation":[{"name":"Wharton School, University of Pennsylvania, Philadelphia, Pennsylvania 19104;"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/2.zoppoz.workers.dev:443\/https\/orcid.org\/0000-0002-7280-912X","authenticated-orcid":false,"given":"Mohsen","family":"Bayati","sequence":"additional","affiliation":[{"name":"Stanford Graduate School of Business, Stanford University, Stanford, California 94305;"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/2.zoppoz.workers.dev:443\/https\/orcid.org\/0000-0003-3997-2766","authenticated-orcid":false,"given":"Khashayar","family":"Khosravi","sequence":"additional","affiliation":[{"name":"Stanford University Electrical Engineering, Stanford University, Stanford, California 94305"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"109","reference":[{"key":"B6","author":"Bastani H","year":"2019","journal-title":"Oper. Res."},{"key":"B11","doi-asserted-by":"publisher","DOI":"10.1287\/opre.1120.1057"},{"key":"B13","doi-asserted-by":"publisher","DOI":"10.1214\/aos\/1017938919"},{"key":"B18","doi-asserted-by":"publisher","DOI":"10.1287\/mnsc.2013.1788"},{"key":"B19","first-page":"586","author":"Filippi S","year":"2010","journal-title":"Adv. Neural Inform. Processing Systems"},{"key":"B20","doi-asserted-by":"publisher","DOI":"10.1111\/j.2517-6161.1979.tb01068.x"},{"key":"B21","doi-asserted-by":"publisher","DOI":"10.1214\/08-AAP589"},{"key":"B22","doi-asserted-by":"publisher","DOI":"10.1287\/11-SSY032"},{"key":"B23","first-page":"3153","author":"Gutin E","year":"2016","journal-title":"Adv. Neural Inform. Processing Systems"},{"issue":"1","key":"B25","first-page":"315","volume":"20","author":"Javanmard A","year":"2019","journal-title":"J. Machine Learn. Res."},{"key":"B30","doi-asserted-by":"publisher","DOI":"10.1287\/opre.2014.1294"},{"key":"B32","doi-asserted-by":"publisher","DOI":"10.1158\/2159-8274.CD-10-0010"},{"key":"B33","doi-asserted-by":"publisher","DOI":"10.1016\/0196-8858(85)90002-8"},{"key":"B35","first-page":"550","volume":"27","author":"Lattimore T","year":"2014","journal-title":"Adv. Neural Inform. Processing Systems"},{"key":"B36","volume-title":"Theory of Point Estimation","author":"Lehmann EL","year":"1998"},{"key":"B39","doi-asserted-by":"publisher","DOI":"10.1007\/978-1-4899-3242-6"},{"key":"B40","doi-asserted-by":"publisher","DOI":"10.1109\/TAC.2009.2031725"},{"key":"B42","doi-asserted-by":"publisher","DOI":"10.1080\/00207178708933715"},{"key":"B43","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-319-56393-0"},{"key":"B47","doi-asserted-by":"publisher","DOI":"10.1287\/moor.2014.0650"},{"key":"B48","doi-asserted-by":"publisher","DOI":"10.1214\/aos\/1176348382"},{"key":"B49","doi-asserted-by":"crossref","unstructured":"Tewari A, Murphy SA (2017) From ads to interventions: Contextual bandits in mobile health. Rehg J, Murphy S, Kumar S, eds. Mobile Health (Springer, New York), 495\u2013517.","DOI":"10.1007\/978-3-319-51394-2_25"},{"key":"B50","doi-asserted-by":"publisher","DOI":"10.1093\/biomet\/25.3-4.285"},{"key":"B51","doi-asserted-by":"crossref","unstructured":"Tropp JA (2011) User-friendly tail bounds for matrix martingales. Technical Report TR-2011-01, California Institute of Technology, Pasadena.","DOI":"10.21236\/ADA555817"},{"key":"B52","first-page":"135","author":"Tsybakov AB","year":"2004","journal-title":"Ann. Statist."},{"key":"B53","volume-title":"High-Dimensional Statistics: A Non-Asymptotic Viewpoint,","author":"Wainwright M","year":"2016"},{"key":"B54","doi-asserted-by":"publisher","DOI":"10.1016\/j.aam.2004.10.004"},{"key":"B55","doi-asserted-by":"publisher","DOI":"10.1109\/TAC.2005.844079"},{"key":"B56","doi-asserted-by":"publisher","DOI":"10.1080\/01621459.1979.10481033"},{"key":"B57","unstructured":"Wu Y, Shariff R, Lattimore T, Szepesvari C (2016) Conservative bandits. Balcan MF, Weinberger KQ, eds. Proc. 33rd Internat. Conf. Machine Learn., vol. 48 (JMLR.org, New York), 1254\u20131262."}],"container-title":["Management Science"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/2.zoppoz.workers.dev:443\/https\/pubsonline.informs.org\/doi\/pdf\/10.1287\/mnsc.2020.3605","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2023,4,2]],"date-time":"2023-04-02T11:14:40Z","timestamp":1680434080000},"score":1,"resource":{"primary":{"URL":"https:\/\/2.zoppoz.workers.dev:443\/https\/pubsonline.informs.org\/doi\/10.1287\/mnsc.2020.3605"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2021,3]]},"references-count":30,"journal-issue":{"issue":"3","published-print":{"date-parts":[[2021,3]]}},"alternative-id":["10.1287\/mnsc.2020.3605"],"URL":"https:\/\/2.zoppoz.workers.dev:443\/https\/doi.org\/10.1287\/mnsc.2020.3605","relation":{},"ISSN":["0025-1909","1526-5501"],"issn-type":[{"value":"0025-1909","type":"print"},{"value":"1526-5501","type":"electronic"}],"subject":[],"published":{"date-parts":[[2021,3]]}}}