Presenter: Dr. techn.Dmitriy Shutin, DLR, Oberpfaffenhofen

Inviting Professor: Prof. Dr. Armin Dekorsy

Abstract: 

This talk explores Sparse Bayesian Learning (SBL) as a flexible and powerful approach for identifying simple underlying structure in complex data. After introducing the general idea of sparsity, we show how SBL provides a probabilistic view of sparsity, naturally balancing model complexity and data fit. New computational methods are developed for linear models, with particular attention to distributed settings, where data and computations are spread across multiple agents on the EDGE. A statistical perspective is used to better understand when and why sparse solutions can be reliably recovered. The approach is then extended to nonlinear models, enabling applications to more challenging problems. We show how the method can be used for the localization of gas emission sources governed by physical laws, as well as to line spectral estimation - a key task in signal processing. Together, the results demonstrate how SBL connects statistical inference, efficient computation, and physical modeling to address a wide range of practical problems.