Federated Posterior Sharing for Multi-Agent Systems in Uncertain Environments

Yuxi Wang, Peng Wu, Mahdi Imani
Proceedings of the 7th Annual Learning for Dynamics \& Control Conference, PMLR 283:817-829, 2025.

Abstract

The use of artificial intelligence (AI) agents is increasingly growing in complex, dynamic environments such as disaster response, search and rescue, and law enforcement. These domains are often only partially known, requiring agents to learn and adapt as they gather more information. In multi-agent settings, where agents operate independently and possess diverse, partial views of the environment, sharing their environmental knowledge is essential for enhancing operational efficiency and safety. Existing federated learning approaches focus primarily on policy sharing without modeling environmental uncertainty. To address this gap, this paper presents a framework that enables multiple agents to collaboratively share their probabilistic knowledge of the environment, building a global, shared understanding that efficiently guides their policies. Unlike existing data fusion techniques that exchange raw data–posing privacy risks and increasing communication costs–our method fuses agents’ local posterior distributions as an abstract representation of their past data. We provide both single-step and multi-step synchronization, enabling recursive aggregation of agents’ knowledge to support informed and adaptive decision-making. Numerical experiments show that our method achieves superior accuracy and decision efficiency compared to several existing methods, particularly in settings with heterogeneous priors and greater uncertainty.

Cite this Paper

BibTeX
@InProceedings{pmlr-v283-wang25e, title = {Federated Posterior Sharing for Multi-Agent Systems in Uncertain Environments}, author = {Wang, Yuxi and Wu, Peng and Imani, Mahdi}, booktitle = {Proceedings of the 7th Annual Learning for Dynamics \& Control Conference}, pages = {817--829}, year = {2025}, editor = {Ozay, Necmiye and Balzano, Laura and Panagou, Dimitra and Abate, Alessandro}, volume = {283}, series = {Proceedings of Machine Learning Research}, month = {04--06 Jun}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v283/main/assets/wang25e/wang25e.pdf}, url = {https://proceedings.mlr.press/v283/wang25e.html}, abstract = {The use of artificial intelligence (AI) agents is increasingly growing in complex, dynamic environments such as disaster response, search and rescue, and law enforcement. These domains are often only partially known, requiring agents to learn and adapt as they gather more information. In multi-agent settings, where agents operate independently and possess diverse, partial views of the environment, sharing their environmental knowledge is essential for enhancing operational efficiency and safety. Existing federated learning approaches focus primarily on policy sharing without modeling environmental uncertainty. To address this gap, this paper presents a framework that enables multiple agents to collaboratively share their probabilistic knowledge of the environment, building a global, shared understanding that efficiently guides their policies. Unlike existing data fusion techniques that exchange raw data–posing privacy risks and increasing communication costs–our method fuses agents’ local posterior distributions as an abstract representation of their past data. We provide both single-step and multi-step synchronization, enabling recursive aggregation of agents’ knowledge to support informed and adaptive decision-making. Numerical experiments show that our method achieves superior accuracy and decision efficiency compared to several existing methods, particularly in settings with heterogeneous priors and greater uncertainty.} }
Endnote
%0 Conference Paper %T Federated Posterior Sharing for Multi-Agent Systems in Uncertain Environments %A Yuxi Wang %A Peng Wu %A Mahdi Imani %B Proceedings of the 7th Annual Learning for Dynamics \& Control Conference %C Proceedings of Machine Learning Research %D 2025 %E Necmiye Ozay %E Laura Balzano %E Dimitra Panagou %E Alessandro Abate %F pmlr-v283-wang25e %I PMLR %P 817--829 %U https://proceedings.mlr.press/v283/wang25e.html %V 283 %X The use of artificial intelligence (AI) agents is increasingly growing in complex, dynamic environments such as disaster response, search and rescue, and law enforcement. These domains are often only partially known, requiring agents to learn and adapt as they gather more information. In multi-agent settings, where agents operate independently and possess diverse, partial views of the environment, sharing their environmental knowledge is essential for enhancing operational efficiency and safety. Existing federated learning approaches focus primarily on policy sharing without modeling environmental uncertainty. To address this gap, this paper presents a framework that enables multiple agents to collaboratively share their probabilistic knowledge of the environment, building a global, shared understanding that efficiently guides their policies. Unlike existing data fusion techniques that exchange raw data–posing privacy risks and increasing communication costs–our method fuses agents’ local posterior distributions as an abstract representation of their past data. We provide both single-step and multi-step synchronization, enabling recursive aggregation of agents’ knowledge to support informed and adaptive decision-making. Numerical experiments show that our method achieves superior accuracy and decision efficiency compared to several existing methods, particularly in settings with heterogeneous priors and greater uncertainty.
APA
Wang, Y., Wu, P. & Imani, M.. (2025). Federated Posterior Sharing for Multi-Agent Systems in Uncertain Environments. Proceedings of the 7th Annual Learning for Dynamics \& Control Conference, in Proceedings of Machine Learning Research 283:817-829 Available from https://proceedings.mlr.press/v283/wang25e.html.

Related Material