A globally convergent algorithm for nonconvex optimization based on block coordinate update (Q1676921)
From MaRDI portal
scientific article
| Language | Label | Description | Also known as |
|---|---|---|---|
| English | A globally convergent algorithm for nonconvex optimization based on block coordinate update |
scientific article |
Statements
A globally convergent algorithm for nonconvex optimization based on block coordinate update (English)
0 references
10 November 2017
0 references
A block prox-linear (BPL) method for nonconvex smooth and nonsmooth optimization is presented. Assuming cyclic updates of the blocks it is proved that the whole sequence generated by the BPL method converges to a critical point (global convergence). Its asymptotic convergence rate is given, too. The efficiency of this method is numerically demonstrated by numerical tests on nonnegative matrices and tensor factorization problems.
0 references
nonconvex optimization
0 references
nonsmooth optimization
0 references
block coordinate descent
0 references
Kurdyka-Łojasiewicz inequality
0 references
prox-linear
0 references
whole sequence convergence
0 references
global convergence
0 references
asymptotic convergence
0 references
numerical test
0 references
0 references
0 references
0 references
0 references
0 references
0 references
0 references
0 references
0 references
0 references
