Exploration–Exploitation in MDPs with Options - Archive ouverte HAL Access content directly
Conference Papers Year :

Exploration–Exploitation in MDPs with Options

(1, 2) , (1, 2)
1
2

Abstract

While a large body of empirical results show that temporally-extended actions and options may significantly affect the learning performance of an agent, the theoretical understanding of how and when options can be beneficial in online reinforcement learning is relatively limited. In this paper, we derive an upper and lower bound on the regret of a variant of UCRL using options. While we first analyze the algorithm in the general case of semi-Markov decision processes (SMDPs), we show how these results can be translated to the specific case of MDPs with options and we illustrate simple scenarios in which the regret of learning with options can be provably much smaller than the regret suffered when learning with primitive actions.
Fichier principal
Vignette du fichier
main.pdf (626.88 Ko) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-01493567 , version 1 (21-03-2017)
hal-01493567 , version 2 (24-03-2017)

Identifiers

  • HAL Id : hal-01493567 , version 2

Cite

Ronan Fruit, Alessandro Lazaric. Exploration–Exploitation in MDPs with Options. AISTATS 2017 - 20th International Conference on Artificial Intelligence and Statistics, Apr 2017, Fort Lauderdale, United States. ⟨hal-01493567v2⟩
222 View
136 Download

Share

Gmail Facebook Twitter LinkedIn More