Finite Time Analysis of Linear Two-timescale Stochastic Approximation with Markovian Noise - Archive ouverte HAL Access content directly
Conference Papers Year :

Finite Time Analysis of Linear Two-timescale Stochastic Approximation with Markovian Noise

(1) , (2, 1, 3) , (1) , (4) , (5)
1
2
3
4
5

Abstract

Linear two-timescale stochastic approximation (SA) scheme is an important class of algorithms which has become popular in reinforcement learning (RL), particularly for the policy evaluation problem. Recently, a number of works have been devoted to establishing the finite time analysis of the scheme, especially under the Markovian (non-i.i.d.) noise settings that are ubiquitous in practice. In this paper, we provide a finite-time analysis for linear two timescale SA. Our bounds show that there is no discrepancy in the convergence rate between Markovian and martingale noise, only the constants are affected by the mixing time of the Markov chain. With an appropriate step size schedule, the transient term in the expected error bound is o(1/k c) and the steady-state term is O(1/k), where c > 1 and k is the iteration number. Furthermore, we present an asymptotic expansion of the expected error with a matching lower bound of Ω(1/k). A simple numerical experiment is presented to support our theory.
Fichier principal
Vignette du fichier
Finite Time Analysis of Linear Two-timescale Stochastic Approximation with Markovian Noise.pdf (1.28 Mo) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03033458 , version 1 (01-12-2020)

Identifiers

  • HAL Id : hal-03033458 , version 1

Cite

Maxim Kaledin, Eric Moulines, Alexey Naumov, Vladislav Tadic, Hoi-To Wai. Finite Time Analysis of Linear Two-timescale Stochastic Approximation with Markovian Noise. COLT 2020 - 33rd Conference on Learning Theory, Jul 2020, Graz / Virtual, Austria. ⟨hal-03033458⟩
28 View
150 Download

Share

Gmail Facebook Twitter LinkedIn More