Dynamic Diffusion Estimation in Exponential Family Models
Kamil Dedecius, Vladimíra Sečkárová · IEEE Signal Processing Letters · 2013
This letter proposes a new dynamic diffusion estimation method for a collaborative inference of a common model parameter using a distributed network of cooperating nodes. Unlike the existing single problem-oriented diffusion methods, it is formulated abstractly for the exponential family of models. The resulting advantage-its easy and straightforward application to the family members-is demonstrated on three selected cases: the diffusion autoregression, the diffusion Poisson modelling and the diffusion estimation of a Bernoulli process with unknown proportions. The first case is shown to coincide with the diffusion recursive least squares.