Factor Graphs with NUV Priors and Iteratively Reweighted Descent for Sparse Least Squares and More

Model-Based Separation, Detection, and Classification of Eye Movements

“…. , L, are commonly modeled by iid sparsity-promoting priors p(u k ) as described in the Appendix D. Sparse Bayesian learning [36] relies on the hierarchical representation [37] p(u k ) = sup…”

“…with a suitable hyperprior p(σ k ) (see Appendix D, (79)). Since σ 2 k is a variance, such representations have been called normal with unknown variances (NUV) [24], [37]. Using this NUV representation for U Sacc k , for fixed σ Sacc k , the sparse inputs become Gaussian, i.e., U Sacc k ∼ N (0, (σ Sacc k ) 2 ).…”

“…which in the sparse Bayesian learning (SBL) framework [36] are represented via the hierarchical representation [37] p(u k ) = sup…”

Binary Control and Digital-to-Analog Conversion Using Composite NUV Priors and Iterative Gaussian Message Passing

“…The heart of the method proposed in this paper is a new binary-enforcing NUV prior (normal with unknown variance). NUV priors are a central idea of sparse Bayesian learning [9,10,11,12], and closely related to variational representations of Lp norms [13,14]. Such priors have been used mainly for sparsity; in particular, no binary-enforcing NUV prior seems to have been proposed in the prior literature.…”

“…In(13), ρ is used for two different functions (with different arguments).3.3.1. Iterative Kalman Input Estimation (IKIE)The algorithm estimates θ and u by alternating the following two steps for i = 1, 2, 3, .…”