# 42839

## Автор(ов):

1

Параметры публикации

Тезисы доклада

## Название:

Adaptive Mirror Descent Algorithms for a Convex Stochastic Optimization and Multi-Armed Bandit Problems

Да

## Наименование конференции:

• Monash Probability Conference in Honor of Robert Liptser’s 80th Birthday (Prato, Italy, 2016)

## Наименование источника:

• Abstract Book of Monash Probability Conference in Honor of Robert Liptser’s 80th Birthday

• Prato

## Издательство:

• Издательство Monash University of Prato

2016

## Страницы:

22-22
Аннотация
We study the adaptive Mirror Descent Algorithms for a convex stochastic optimization and multi-armed bandit problems. The recursive algorithms realize primal-dual method of gradient type under different parameters like norms in primal/dual spaces, proxi-function that generates a related projection by mapping the dual space onto the given convex optimization set using Legendre-Fenchel transform and the gradient, and two real sequences as step size and a so called generalized temperature. The latter is recursively updated by using current observations (i.e., stochastic sub-gradients for optimizing the function) and leads the algorithms to optimal rate of convergence (up to log-factor). The algorithms demonstrate their effectiveness on the example of the multi-armed bandit problem.

## Библиографическая ссылка:

Назин А.В. Adaptive Mirror Descent Algorithms for a Convex Stochastic Optimization and Multi-Armed Bandit Problems / Abstract Book of Monash Probability Conference in Honor of Robert Liptser’s 80th Birthday. Prato: Издательство Monash University of Prato, 2016. С. 22-22.