Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.649804
Title: Bandit problems on parallel machines
Author: Dunn, R. T.
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 2005
Availability of Full Text:
Full text unavailable from EThOS. Please contact the current institution’s library for further details.
Abstract:
Glazebrook and Wilkinson (2000) exploited the primal-dual approach within achievable region methodology to investigate the classical multi-armed bandit problem on identical machines working in parallel. We follow this analysis by utilising and developing elements of the account, given by Bertsimas and Niño-Mora (1996), of the branching bandit model on a single machine cast in the achievable region framework in order to extend Glazebrook and Wilkinson's work to more general models in the parallel machine environment. We obtain performance guarantees for a range of simple index based heuristic policies for models in which bandits compete for processing by machines of differing speeds and also where the number of available (identical) machines is a stochastic process. Prom these performance guarantees various forms of asymptotic optimality are established. Numerical studies allow insights concerning the degree of conservatism in the theoretical results.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.649804  DOI: Not available
Share: