Diffusion Estimation Of State-Space Models: Bayesian Formulation

0431804 - ÚTIA 2015 RIV FR eng C - Konferenční příspěvek (zahraniční konf.)
Dedecius, Kamil
Diffusion Estimation Of State-Space Models: Bayesian Formulation.
Proceedings of the 24th IEEE International Workshop on Machine Learning for Signal Processing (MLSP2014). Reims: IEEE, 2014. ISBN 978-1-4799-3693-9.
[The 24th IEEE International Workshop on Machine Learning for Signal Processing (MLSP2014). Reims (FR), 21.09.2014-24.09.2014]
Grant CEP: GA ČR(CZ) GP14-06678P
Klíčová slova: distributed estimation * state-space models * Bayesian estimation
Obor OECD: Statistics and probability
http://library.utia.cas.cz/separaty/2014/AS/dedecius-0431804.pdf

The paper studies the problem of decentralized distributed estimation of the state-space models from the Bayesian viewpoint. The adopted diffusion strategy, consisting of collective adaptation to new data and combination of posterior estimates, is derived in general model-independent form. Its particular application to the celebrated Kalman filter demonstrates the ease of use, especially when the measurement model is rewritten into the exponential family form and a conjugate prior describes the estimated state.
Trvalý link: http://hdl.handle.net/11104/0237640