English

Bandit Convex Optimisation Revisited: FTRL Achieves $\tilde{O}(t^{1/2})$ Regret

Machine Learning 2023-06-27 v2

Abstract

We show that a kernel estimator using multiple function evaluations can be easily converted into a sampling-based bandit estimator with expectation equal to the original kernel estimate. Plugging such a bandit estimator into the standard FTRL algorithm yields a bandit convex optimisation algorithm that achieves O~(t1/2)\tilde{O}(t^{1/2}) regret against adversarial time-varying convex loss functions.

Keywords

Cite

@article{arxiv.2302.00358,
  title  = {Bandit Convex Optimisation Revisited: FTRL Achieves $\tilde{O}(t^{1/2})$ Regret},
  author = {David Young and Douglas Leith and George Iosifidis},
  journal= {arXiv preprint arXiv:2302.00358},
  year   = {2023}
}

Comments

Error in proof