Rate optimal learning of equilibria from data

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Freihaut, Till, Viano, Luca, Nevali, Emanuele, Cevher, Volkan, Geist, Matthieu, Ramponi, Giorgia
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912640907870208
author Freihaut, Till
Viano, Luca
Nevali, Emanuele
Cevher, Volkan
Geist, Matthieu
Ramponi, Giorgia
author_facet Freihaut, Till
Viano, Luca
Nevali, Emanuele
Cevher, Volkan
Geist, Matthieu
Ramponi, Giorgia
contents We close open theoretical gaps in Multi-Agent Imitation Learning (MAIL) by characterizing the limits of non-interactive MAIL and presenting the first interactive algorithm with near-optimal sample complexity. In the non-interactive setting, we prove a statistical lower bound that identifies the all-policy deviation concentrability coefficient as the fundamental complexity measure, and we show that Behavior Cloning (BC) is rate-optimal. For the interactive setting, we introduce a framework that combines reward-free reinforcement learning with interactive MAIL and instantiate it with an algorithm, MAIL-WARM. It improves the best previously known sample complexity from $\mathcal{O}(\varepsilon^{-8})$ to $\mathcal{O}(\varepsilon^{-2}),$ matching the dependence on $\varepsilon$ implied by our lower bound. Finally, we provide numerical results that support our theory and illustrate, in environments such as grid worlds, where Behavior Cloning fails to learn.
format Preprint
id arxiv_https___arxiv_org_abs_2510_09325
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Rate optimal learning of equilibria from data
Freihaut, Till
Viano, Luca
Nevali, Emanuele
Cevher, Volkan
Geist, Matthieu
Ramponi, Giorgia
Machine Learning
Artificial Intelligence
We close open theoretical gaps in Multi-Agent Imitation Learning (MAIL) by characterizing the limits of non-interactive MAIL and presenting the first interactive algorithm with near-optimal sample complexity. In the non-interactive setting, we prove a statistical lower bound that identifies the all-policy deviation concentrability coefficient as the fundamental complexity measure, and we show that Behavior Cloning (BC) is rate-optimal. For the interactive setting, we introduce a framework that combines reward-free reinforcement learning with interactive MAIL and instantiate it with an algorithm, MAIL-WARM. It improves the best previously known sample complexity from $\mathcal{O}(\varepsilon^{-8})$ to $\mathcal{O}(\varepsilon^{-2}),$ matching the dependence on $\varepsilon$ implied by our lower bound. Finally, we provide numerical results that support our theory and illustrate, in environments such as grid worlds, where Behavior Cloning fails to learn.
title Rate optimal learning of equilibria from data
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2510.09325