Distributed algorithms reduce communication in multi-agent estimation
Resource-Efficient Distributed Recursive Gaussian Processes
Machine Learning
Summary
When multiple agents gather data to estimate something unknown, combining their information into one shared model can be hard without a central controller. The authors developed two methods that let these agents work together using only local data and chatting with neighbors. These methods keep the shared understanding accurate while cutting down the amount of messages needed. They tested their ideas on real wind data and found their approaches worked well even when the communication connections varied.
What this means in practice
- •For environmental monitoring teams: Enable sensor networks to efficiently combine local readings into accurate, shared environmental models without central control.
- •For industrial automation teams: Improve collaborative state estimation across decentralized machines by reducing communication needs while maintaining accuracy.
Authors
Josephine King, Ali Emre Balci, Raj Thilak Rajan
Abstract
Gaussian processes (GPs) provide a flexible framework for learning unknown functions from noisy measurements while quantifying predictive uncertainty, making them well suited for estimation in multi-agent systems. However, when measurements are collected by multiple agents, maintaining a unified GP model without centralized processing requires efficient distributed algorithms that can operate using local measurements and communication with neighboring agents. In this work, we develop two distributed recursive GP (RGP) algorithms for multi-output GP regression: ADMM-RGP and PDMM-RGP. We analyze the stability and convergence of both algorithms and develop parameter selection strategies to accelerate convergence, thus reducing the communication burden. The proposed methods are validated on a real-world multi-output wind dataset, and their convergence behavior is examined across communication graphs with varying connectivity. Numerical experiments demonstrate that ADMM-RGP and PDMM-RGP can significantly reduce communication relative to the state of the art, while maintaining comparable estimation accuracy and network-wide consensus.