Publication Type
Conference Paper
Book Title
2024 IEEE International Conference on Big Data (BigData)
Publication Date
Page Numbers
4257 to 4264
Publisher Location
New Jersey, United States of America
Conference Name
2024 IEEE International Conference on Big Data (BigData)
Conference Location
Washington DC, District of Columbia, United States of America
Conference Sponsor
NSF, Virginia Tech
Conference Date
-
Abstract
In the compression of scientific data, error-controlled compressors enable to considerably decrease the size of the dataset while maintaining adequate levels of accuracy. In this paper, we note that multi-level refactoring scheme such as MGARD i) rely on an approximation of the data based on the interpolation of coefficients, ii) estimate the resulting error with global metrics on the dataset. To improve on these two aspects, we propose a method that aims to divide the original dataset into blocks based on their smoothness and refactors each block separately with the most relevant interpolation order. We show the relevance of such a method on tailored datasets and the benefits and challenges when applying it to large scientific data.
