Anderson acceleration for geometry optimization and physics simulation
- Submitting institution
-
Cardiff University / Prifysgol Caerdydd
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 96926748
- Type
- D - Journal article
- DOI
-
10.1145/3197517.3201290
- Title of journal
- ACM Transactions on Graphics
- Article number
- 42
- First page
- -
- Volume
- 37
- Issue
- 4
- ISSN
- 0730-0301
- Open access status
- Compliant
- Month of publication
- July
- Year of publication
- 2018
- URL
-
http://dx.doi.org/10.1145/3197517.3201290
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
5
- Research group(s)
-
V - Visual computing
- Citation count
- 6
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- This paper accelerates a class of first-order solvers commonly used for numerical optimisation problems in computer graphics. Such solvers can quickly converge to an approximate solution but can take many more iterations to converge to a high-accuracy solution. Our method is based on Anderson Acceleration - a well-established numerical technique - but with a new stabilisation strategy to avoid stagnation (a well-known issue of Anderson Acceleration). The method was tested on a variety of graphics problems and can reduce the computational time by >90%. The work was presented as a technical paper in SIGGRAPH 2018. Code is available at https://github.com/bldeng/AASolver.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -