Some properties of multidimensional stochastic optimization algorithm with a constant step-size

Yury Ivanskiy, Kseniia Kopylova

Keywords: stochastic optimization, SPSA, finite time convergence, formation control

Abstract:

The paper studies the convergence properties of a zero-order multidimensional stochastic optimization algorithm with a constant step-size.
An estimate of the finite time convergence error is given.
The operation of the algorithm is illustrated by the simulation of the spacecraft spatial position adjustment within a satellite constellation, which is relevant in controlling the orientation of an interferometric mega-constellation space telescope under conditions of uncertainty.

Submission Number: 81