English

Improved Exploiting Higher Order Smoothness in Derivative-free Optimization and Continuous Bandit

Optimization and Control 2021-04-30 v2

Abstract

We consider β\beta-smooth (satisfies the generalized Holder condition with parameter β>2\beta > 2) stochastic convex optimization problem with zero-order one-point oracle. The best known result was arXiv:2006.07862: E[f(xN)f(x)]=O~(n2γNβ1β)\mathbb{E} \left[f(\overline{x}_N) - f(x^*)\right] = \tilde{\mathcal{O}} \left(\dfrac{n^{2}}{\gamma N^{\frac{\beta-1}{\beta}}} \right) in γ\gamma-strongly convex case, where nn is the dimension. In this paper we improve this bound: E[f(xN)f(x)]=O~(n21βγNβ1β).\mathbb{E} \left[f(\overline{x}_N) - f(x^*)\right] = \tilde{\mathcal{O}} \left(\dfrac{n^{2-\frac{1}{\beta}}}{\gamma N^{\frac{\beta-1}{\beta}}} \right).

Keywords

Cite

@article{arxiv.2101.03821,
  title  = {Improved Exploiting Higher Order Smoothness in Derivative-free Optimization and Continuous Bandit},
  author = {Vasilii Novitskii and Alexander Gasnikov},
  journal= {arXiv preprint arXiv:2101.03821},
  year   = {2021}
}
R2 v1 2026-06-23T21:59:06.590Z