Error Forgetting of Bregman Iteration
| dc.contributor.author | Yin, Wotao | |
| dc.contributor.author | Osher, Stanley | |
| dc.date.accessioned | 2018-06-19T17:47:59Z | |
| dc.date.available | 2018-06-19T17:47:59Z | |
| dc.date.issued | 2012-01 | |
| dc.date.note | January 2012 | |
| dc.description.abstract | This short article analyzes an interesting property of the Bregman iterative procedure for minimizing a convex piece-wise linear function J(x) subject to linear constraints Ax=b. The procedure obtains its solution by solving a sequence of unconstrained subproblems, each minimizing J(x) + (1/2) ||Ax-bk||2, and iteratively updating bk. In practice, the subproblems are solved with finite accuracy. Let wk denote the numerical error at iteration k. If all wk are sufficiently small, Bregman iteration identifies the optimal face in finitely many iterations, and afterward, it enjoys an interesting error-forgetting property: the distance between the current point and the optimal solution set is bounded by ||wk+1-wk||, independent of the numerical errors at previous iterations. This property partially explains why the Bregman iterative procedure works well for sparse optimization and ||x||1 minimization. The error-forgetting property is unique to piece-wise linear functions (i.e., polyhedral functions) J(x), and it is new to the literature of the augmented Lagrangian method. | |
| dc.format.extent | 9 pp | |
| dc.identifier.citation | Yin, Wotao and Osher, Stanley. "Error Forgetting of Bregman Iteration." (2012) <a href="https://hdl.handle.net/1911/102193">https://hdl.handle.net/1911/102193</a>. | |
| dc.identifier.digital | TR12-03 | |
| dc.identifier.uri | https://hdl.handle.net/1911/102193 | |
| dc.language.iso | eng | |
| dc.title | Error Forgetting of Bregman Iteration | |
| dc.type | Technical report | |
| dc.type.dcmi | Text |
Files
Original bundle
1 - 1 of 1