Summary
IEICE Information and Communication Technology Forum
2014
Session Number:ICT2
Session:
Number:ICT2-4
Graphical models and message passing in compressed sensing
Aleksandra Pizurica,
pp.-
Publication Date:2014-08-10
Online ISSN:2188-5079
Summary:
This talk reviews some ideas and practical examples of using graphical models and message passing for the reconstruction of sparse signals. Graphical models offer an elegant way to encode signal structure. We discuss a particular case of using Markov Random Field (MRF) models to improve sparse reconstruction in Magnetic Resonance Imaging (MRI). Another aspect of using graphical models in compressed sensing is designing fast iterative reconstruction algorithms based on belief propagation and related message passing principles. We discuss a class of iterative recovery algorithms inspired by low-density parity check (LDPC) coding and the so-called approximate message passing (AMP) algorithms in compressed sensing.