Sparsity-Enabled Step Width Adaption for Linearized Bregman based Algorithms
Sprache des Vortragstitels:
IEEE Statistical Signal Processing Workshop (SSP 2018)
Sprache des Tagungstitel:
Iterative algorithms based on linearized Bregman iterations allow ef?ciently solving sparse estimation problems. Especially the Kaczmarz and sparse least mean squares ?lter (LMS) variants are very suitable for implementation in digital hard- and software. However, when analyzing the error of such algorithms over the iterations one realizes that especially at early iterations only small error reductions occur. To improve this behavior, we propose to use sparsity-enabled step width adaption. We show simulations results demonstrating that this approach signi?cantly improves the performance of sparse Kaczmarz and sparse LMS algorithms.