The Coordinate-Descent-Type Methods
Randomized Kaczmarz Iteration Methods:Algorithmic Extensions and Convergence Theory.
Part II: The Coordinate-Descent-Type Methods
We review and compare several representative and effective randomized coordinate-descent-type methods, and their modifications and extensions, for solving the large, sparse, consistent or inconsistent systems of linear equations. We also anatomize, extract, and purify the asymptotic convergence theories of these iteration methods, and discuss, analyze, and summarize their advantages and disadvantages from the viewpoints of both theory and computations.