EUVIMEDOpdag EUVIMED
Open access · CC BY 4.0

Explainable Reinforcement Learning Regulators for Best-Response Dynamics in Shapley Facility Location Games

Andres Torres-Reyes, Rolando Menchaca-Méndez

Mathematics · 2026-11-14 · 14 (22) · 4186

DOI: 10.3390/math14224186 ↗

At a glance

Journal article in Mathematics, published online on 2026-11-14. Written by Andres Torres-Reyes and 1 co-authors. 53 references. The full text is freely available (open access, CC BY 4.0). Funded by: Secretaría de Ciencia, Humanidades, Tecnología e Innovación, Instituto Politécnico Nacional.

AbstractBibliografiske oplysningerCiteRelated studies

Abstract (original language)

Asynchronous best-response dynamics in finite exact potential games converge to pure Nash equilibria, but total computational cost depends on player activation schedules. We formulate activation scheduling as a Markov decision process, in which a central regulator learns, with tabular Q-learning, which player to activate in a discrete Shapley facility location game. To handle the combinatorial state space, we present a permutation-invariant observation mapping with four binary flags per player that compresses state representations from mn to n+15n. This reduces per-step decision costs to Θ(m2n), compared to Θ(m2n2) for greedy selection. We prove that the regulator asymptotically dominates greedy selection in total cost if and only if their expected iteration ratio is o(n), and empirically achieve up to a 2.91-fold cost reduction. A hand-crafted baseline, lowest payoff first, achieves costs 0.34–0.65 times the regulator’s by exploiting an anti-correlation between current payoffs and maximum available gains. The learned Q-tables show the same preference, drawn from the single payoff bit they observe. Finally, restricting the learned choice to unchecked players removes the stalls caused by state aliasing, and a pilot distills the policy into interpretable decision trees that provably reach equilibrium.

Funding

  • Secretaría de Ciencia, Humanidades, Tecnología e Innovación
  • Instituto Politécnico Nacional

Cite

APA

Torres-Reyes, A., & Menchaca-Méndez, R. (2026). Explainable Reinforcement Learning Regulators for Best-Response Dynamics in Shapley Facility Location Games. Mathematics, 14(22), 4186. https://doi.org/10.3390/math14224186

Vancouver

Torres-Reyes A, Menchaca-Méndez R. Explainable Reinforcement Learning Regulators for Best-Response Dynamics in Shapley Facility Location Games. Mathematics. 2026;14(22):4186. doi:10.3390/math14224186

BibTeX
@article{TorresReyes2026,
  author = {Torres-Reyes, Andres and Menchaca-Méndez, Rolando},
  title = {Explainable Reinforcement Learning Regulators for Best-Response Dynamics in Shapley Facility Location Games},
  journal = {Mathematics},
  year = {2026},
  volume = {14},
  number = {22},
  pages = {4186},
  doi = {10.3390/math14224186}
}
Start for free →