English

Full Gradient Deep Reinforcement Learning for Average-Reward Criterion

Systems and Control 2023-04-10 v1 Machine Learning Systems and Control

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.

Cite

@article{arxiv.2304.03729,
  title  = {Full Gradient Deep Reinforcement Learning for Average-Reward Criterion},
  author = {Tejas Pagare and Vivek Borkar and Konstantin Avrachenkov},
  journal= {arXiv preprint arXiv:2304.03729},
  year   = {2023}
}

Comments

13 pages, 4 figures; Accepted by 5th Annual Learning for Dynamics & Control Conference (L4DC) 2023

R2 v1 2026-06-28T09:54:41.151Z