RUS  ENG
Full version
JOURNALS // Problemy Peredachi Informatsii // Archive

Probl. Peredachi Inf., 2020 Volume 56, Issue 3, Pages 86–111 (Mi ppi2323)

This article is cited in 9 papers

Automata Theory

Gaussian two-armed bandit: limiting description

A. V. Kolnogorov

Department of Applied Mathematics and Information Science, Yaroslav-the-Wise Novgorod State University, Novgorod, Russia

Abstract: For a Gaussian two-armed bandit, which arises when batch data processing is analyzed, the minimax risk limiting behavior is investigated as the control horizon $N$ grows infinitely. The minimax risk is searched for as the Bayesian one computed with respect to the worst-case prior distribution. We show that the highest requirements are imposed on the control in the domain of “close” distributions where mathematical expectations of incomes differ by a quantity of the order of $N^{-1/2}$. In the domain of “close” distributions, we obtain a recursive integro-difference equation for finding the Bayesian risk with respect to the worst-case prior distribution, in invariant form with control horizon one, and also a second-order partial differential equation in the limiting case. The results allow us to estimate the performance of batch processing. For example, the minimax risk corresponding to batch processing of data partitioned into $50$ batches can be only $2\%$ greater than its limiting value when the number of batches grows infinitely. In the case of a Bernoulli two-armed bandit, we show that optimal one-by-one data processing is not more efficient than batch processing as $N$ grows infinitely.

Keywords: Gaussian two-armed bandit, minimax and Bayesian approaches, batch processing, asymptotic minimax theorem.

UDC: 621.391.1 : 519.713 : 517.977.5

Received: 06.04.2020
Revised: 02.06.2020
Accepted: 02.06.2020

DOI: 10.31857/S0555292320030055


 English version:
Problems of Information Transmission, 2020, 56:3, 278–301

Bibliographic databases:


© Steklov Math. Inst. of RAS, 2024