Full Gradient Deep Reinforcement Learning for Average-Reward Criterion - Inria - Institut national de recherche en sciences et technologies du numérique Access content directly
Conference Papers Year : 2023

Full Gradient Deep Reinforcement Learning for Average-Reward Criterion

Tejas Pagare
  • Function : Author
  • PersonId : 1331389
Vivek S Borkar
  • Function : Author
  • PersonId : 913857

Abstract

We extend the provably convergent Full Gradient DQN algorithm for discounted reward Markov decision processes from Avrachenkov et al. (2021) to average reward problems. We experimentally compare widely used RVI Q-learning with recently proposed Differential Q-learning in the neural function approximation setting with Full Gradient DQN and DQN. We also extend this to learn Whittle indices for Markovian restless multi-armed bandits. We observe a better convergence rate of the proposed Full Gradient variant across different tasks.
Fichier principal
Vignette du fichier
PagareL4DC23.pdf (1.34 Mo) Télécharger le fichier
Origin : Publisher files allowed on an open archive
Licence : CC BY - Attribution

Dates and versions

hal-04372096 , version 1 (04-01-2024)

Licence

Attribution

Identifiers

  • HAL Id : hal-04372096 , version 1

Cite

Tejas Pagare, Vivek S Borkar, Konstantin Avrachenkov. Full Gradient Deep Reinforcement Learning for Average-Reward Criterion. L4DC - The 5th Annual Learning for Dynamics and Control Conference, Jun 2023, Philadelphia, United States. pp.235-247. ⟨hal-04372096⟩
18 View
10 Download

Share

Gmail Facebook X LinkedIn More