Anderson Acceleration for Nonconvex ADMM Based on Douglas-Rachford Splitting

dc.contributor.authorOuyang, Wenqingen_US
dc.contributor.authorPeng, Yueen_US
dc.contributor.authorYao, Yuxinen_US
dc.contributor.authorZhang, Juyongen_US
dc.contributor.authorDeng, Bailinen_US
dc.contributor.editorJacobson, Alec and Huang, Qixingen_US
dc.date.accessioned2020-07-05T13:26:19Z
dc.date.available2020-07-05T13:26:19Z
dc.date.issued2020
dc.description.abstractThe alternating direction multiplier method (ADMM) is widely used in computer graphics for solving optimization problems that can be nonsmooth and nonconvex. It converges quickly to an approximate solution, but can take a long time to converge to a solution of high-accuracy. Previously, Anderson acceleration has been applied to ADMM, by treating it as a fixed-point iteration for the concatenation of the dual variables and a subset of the primal variables. In this paper, we note that the equivalence between ADMM and Douglas-Rachford splitting reveals that ADMM is in fact a fixed-point iteration in a lower-dimensional space. By applying Anderson acceleration to such lower-dimensional fixed-point iteration, we obtain a more effective approach for accelerating ADMM. We analyze the convergence of the proposed acceleration method on nonconvex problems, and verify its effectiveness on a variety of computer graphics including geometry processing and physical simulation.en_US
dc.description.number5
dc.description.sectionheadersOptimization
dc.description.seriesinformationComputer Graphics Forum
dc.description.volume39
dc.identifier.doi10.1111/cgf.14081
dc.identifier.issn1467-8659
dc.identifier.pages221-239
dc.identifier.urihttps://doi.org/10.1111/cgf.14081
dc.identifier.urihttps://diglib.eg.org:443/handle/10.1111/cgf14081
dc.publisherThe Eurographics Association and John Wiley & Sons Ltd.en_US
dc.rightsAttribution 4.0 International License
dc.rights.urihttps://creativecommons.org/licenses/by/4.0/
dc.subjectMathematics of computing
dc.subjectSolvers
dc.subjectMathematical optimization
dc.subjectNumerical analysis
dc.titleAnderson Acceleration for Nonconvex ADMM Based on Douglas-Rachford Splittingen_US
Files
Original bundle
Now showing 1 - 2 of 2
Loading...
Thumbnail Image
Name:
v39i5pp221-239.pdf
Size:
4.26 MB
Format:
Adobe Portable Document Format
No Thumbnail Available
Name:
code.zip
Size:
60.71 MB
Format:
Zip file
Collections