1: \begin{abstract}

2: In this paper, we propose an inexact block coordinate descent algorithm

3: for large-scale nonsmooth nonconvex optimization problems. At each

4: iteration, a particular block variable is selected and updated by

5: inexactly solving the original optimization problem with respect to

6: that block variable. More precisely, a local approximation of the

7: original optimization problem is solved. The proposed algorithm has

8: several attractive features, namely, i) high flexibility, as the approximation

9: function only needs to be strictly convex and it does not have to

10: be a global upper bound of the original function; ii) fast convergence,

11: as the approximation function can be designed to exploit the problem

12: structure at hand and the stepsize is calculated by the line search;

13: iii) low complexity, as the approximation subproblems are much easier

14: to solve and the line search scheme is carried out over a properly

15: constructed differentiable function; iv) guaranteed convergence of

16: a subsequence to a stationary point, even when the objective function

17: does not have a Lipschitz continuous gradient. Interestingly, when

18: the approximation subproblem is solved by a descent algorithm, convergence

19: of a subsequence to a stationary point is still guaranteed even if

20: the approximation subproblem is solved inexactly by terminating the

21: descent algorithm after a finite number of iterations. These features

22: make the proposed algorithm suitable for large-scale problems where

23: the dimension exceeds the memory and/or the processing capability

24: of the existing hardware. These features are also illustrated by several

25: applications in signal processing and machine learning, for instance,

26: network anomaly detection and phase retrieval.

27: \end{abstract}