Distributed Dynamic Estimation in Diffusion Networks

DiffEst - Project Annotation

The project aims to develop a new dynamic estimation framework, intended for spatially distributed low-cost estimation of stationary and nonstationary processes. Being designed for diffusion networks, where each node can use information provided by neighbour nodes, it will not rely on the existence of a dedicated fusion center, nor a Hamiltonian cycle. The framework will be formulated abstractly in the Bayesian paradigm, allowing, in contrast to current single-problem-oriented methods, its direct application to a large class of different problems, comprising dynamic distributed (auto)regression, classification, reliability estimation etc. The developed methods will be efficient in terms of computational, communication and energy resources. Their robustness to network elements degradation and failures is an inherent part of the solution.

In 2014-2017, the works were supported by the Czech Science Foundation under project id GP14-06678P.