Coarse-Grained Smoothness for Reinforcement Learning in Metric Spaces

Abstract

Principled decision-making in continuous state–action spaces is impossible without some assumptions. A common approach is to assume Lipschitz continuity of the Q-function. We show that, unfortunately, this property fails to hold in many typical domains. We propose a new coarse-grained smoothness definition that generalizes the notion of Lipschitz continuity, is more widely applicable, and allows us to compute significantly tighter bounds on Q-functions, leading to improved learning. We provide a theoretical analysis of our new smoothness definition, and discuss its implications and impact on control and exploration in continuous domains.

Cite

Text

Gottesman et al. "Coarse-Grained Smoothness for Reinforcement Learning in Metric Spaces." Artificial Intelligence and Statistics, 2023.

Markdown

[Gottesman et al. "Coarse-Grained Smoothness for Reinforcement Learning in Metric Spaces." Artificial Intelligence and Statistics, 2023.](https://mlanthology.org/aistats/2023/gottesman2023aistats-coarsegrained/)

BibTeX

@inproceedings{gottesman2023aistats-coarsegrained,
  title     = {{Coarse-Grained Smoothness for Reinforcement Learning in Metric Spaces}},
  author    = {Gottesman, Omer and Asadi, Kavosh and Allen, Cameron S. and Lobel, Samuel and Konidaris, George and Littman, Michael},
  booktitle = {Artificial Intelligence and Statistics},
  year      = {2023},
  pages     = {1390-1410},
  volume    = {206},
  url       = {https://mlanthology.org/aistats/2023/gottesman2023aistats-coarsegrained/}
}