On the dynamics of multi agent nonlinear filtering and learning

Multiagent systems aim to accomplish highly complex learning tasks through decentralised consensus seeking dynamics. In the past decade, use of the multiagent framework either in the form of federated learning or distributed estimation has garnered a great deal of attention in the signal processing and computational intelligence societies. This article examines behaviour of multiagent networked systems that use nonlinear filtering/learning frameworks. This article considers the most general learning framework that is reliant on multiple input streams. Operations of this centralised framework is decomposed into operations corresponding to the federated and distributed frameworks. This decomposition process is used to formulate the changes in learning dynamics as the framework shifts from centralised to distributed. Importantly, the formulated dynamics indicate convergence conditions of distributed learning techniques are much more relaxed than formerly considered.