Oxford Events, the new replacement for OxTalks, will launch on 16th March. From now until the launch of Oxford Events, new events cannot be published or edited on OxTalks while all existing records are migrated to the new platform. The existing OxTalks site will remain available to view during this period.
From 16th, Oxford Events will launch on a new website: events.ox.ac.uk, and event submissions will resume. You will need a Halo login to submit events. Full details are available on the Staff Gateway.
Narendra-Shapiro (NS) algorithms are bandit-type algorithms introduced in the sixties (with a view to applications in Psychology or learning automata), whose convergence has been intensively studied in the stochastic algorithm literature. In this talk, we study the efficiency of these bandit algorithms from a regret point of view. We show that some competitive bounds can be obtained for such algorithms in a modified penalized version. Up to an over-penalization modification, the pseudo-regret Rn related to the penalized two-armed bandit is uniformly bounded by C sqrt(n) (for a known C). We also generalize existing convergence and rates of convergence results to the multi-armed case of the over-penalized bandit algorithm, including the convergence toward the invariant measure of a Piecewise Deterministic Markov Process (PDMP) after a suitable renormalization. Finally, ergodic properties of this PDMP are given in the multi-armed case.
