- Seminar: BLIND AND OFF-GRID ACOUSTIC ECHOES RETRIEVAL USING MULTICHANNEL ANNIHILATING FILTERS
Sparse approximation (also known as compressive sensing) is an active research direction in signal processing. The classical approach is based on convex relaxation: replacement the sparsity constraint with a constraint on the ell-1 norm (sum of the absolute values). There exist fast and robust methods for solving ell-1 regularized approximation problems. This project is inspired by a recent result on the link between sparsity and matrix low-rank approximation/completion (see, http://homepages.vub.ac.be/~imarkovs/publications/ica18b.pdf). Methods and software for this latter approach are available from http://slra.github.io.
The goal of the project is to implement and test on simulated and real data examples the ell-1 norm regularization and low-rank approximation methods.