Inverse Reinforcement Learning for Preference-Aware Coordination in Multi-Agent Decision Processes
American Journal of Data Science and Analysis · 01.04.2026 · 7 (1) · 90-99
DOI: 10.71465/ajdsa3669 ↗Kurz gefasst
Fachartikel in der Zeitschrift American Journal of Data Science and Analysis, online erschienen am 01.04.2026. Verfasst von Liam O’Connor und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
In many collaborative environments, optimal coordination depends on implicit preferences that are not directly observable. This work explores inverse reinforcement learning (IRL) to infer latent reward structures from observed interaction trajectories. A maximum entropy IRL framework is combined with policy optimization to align agent behavior with inferred preferences. The method is validated on 7,800 decision trajectories collected from simulated negotiation and resource allocation environments. Results show that preference-aware policies improve alignment with target outcomes by 21.3% and reduce conflict frequency by 25.7% compared to manually designed reward functions. Furthermore, inferred reward models generalize across similar tasks with less than 10% performance loss. These findings indicate that IRL provides an effective mechanism for capturing hidden objectives in collaborative systems.
Zitieren
APA
O’Connor, L., Williams, M. J., & Smith, D. (2026). Inverse Reinforcement Learning for Preference-Aware Coordination in Multi-Agent Decision Processes. American Journal of Data Science and Analysis, 7(1), 90-99. https://doi.org/10.71465/ajdsa3669
Vancouver
O’Connor L, Williams MJ, Smith D. Inverse Reinforcement Learning for Preference-Aware Coordination in Multi-Agent Decision Processes. American Journal of Data Science and Analysis. 2026;7(1):90-99. doi:10.71465/ajdsa3669
BibTeX
@article{OConnor2026,
author = {O’Connor, Liam and Williams, Matthew J. and Smith, Daniel},
title = {Inverse Reinforcement Learning for Preference-Aware Coordination in Multi-Agent Decision Processes},
journal = {American Journal of Data Science and Analysis},
year = {2026},
volume = {7},
number = {1},
pages = {90--99},
doi = {10.71465/ajdsa3669}
}
EUVIMED