1: \begin{abstract}

2: Coordinate descent algorithms solve optimization problems by

3: successively performing approximate minimization along coordinate

4: directions or coordinate hyperplanes. They have been used in

5: applications for many years, and their popularity continues to grow

6: because of their usefulness in data analysis, machine learning, and

7: other areas of current interest. This paper describes the fundamentals

8: of the coordinate descent approach, together with variants and

9: extensions and their convergence properties, mostly with reference to

10: convex objectives.  We pay particular attention to a certain problem

11: structure that arises frequently in machine learning applications,

12: showing that efficient implementations of accelerated coordinate

13: descent algorithms are possible for problems of this type. We also

14: present some parallel variants and discuss their convergence

15: properties under several models of parallel execution.

16: 

17: \keywords{coordinate descent \and  randomized algorithms \and parallel numerical computing}

18: % \PACS{PACS code1 \and PACS code2 \and more}

19: % \subclass{MSC code1 \and MSC code2 \and more}

20: \end{abstract}

21: