Skip to content

Stochastic optimization research conducted together with Alexander Gasnikov

Notifications You must be signed in to change notification settings

SuperCrabLover/Stochastic-Optimization-For-Zero-Order-Oracle

Repository files navigation

Application of the stochastic smoothing method for solving problems with a zero-order oracle

Introduction

Many approaches to solving optimization problems with zeroth-order oracle are constrained by non-smooth object functions. In this work we are focused on smooth problems with only zero-order information accessible. Such tasks are still viable, since sometimes we can not use autograd for obtaining first order information (art.1). The approach is based on "stohastic smoothing" (art.2) using stohastic gradient for gradient approximation. This article provides iteration and zeroth-order oracle call number bounds.

Poster

All work and obtained results are located in Posted.pdf

Releases

No releases published

Packages

No packages published