LAUSR

Collaborative estimation of partially common parameters over ad hoc diffusion networks where the nodes directly communicate with their neighbors is a challenging task. The problem complexity is significantly high under the lack of knowledge which parameters are shared and among which network nodes. In this paper, we propose an adaptive framework suitable for this task. It is abstractly formulated in the Bayesian and information-theoretic paradigms and, therefore, versatile and easily applicable to a relatively wide class of models. If the observation models belong to the exponential family and the same functional types of prior probability distributions are used for estimation of the shared parameters, the method reduces to an analytically tractable variational algorithm extended by a procedure that passes messages among network nodes. A simulation example demonstrates that the collaboration improves estimation performance of both the shared and strictly local parameters, compared with the noncollaborative scenario.