Practical Shuffle Coding
Abstract
We present a general method for lossless compression of unordered data structures, including multisets and graphs. It is a variant of shuffle coding that is many orders of magnitude faster than the original and enables 'one-shot' compression of single unordered objects. Our method achieves state-of-the-art compression rates on various large-scale network graphs at speeds of megabytes per second, efficiently handling even a multi-gigabyte plain graph with one billion edges. We release an implementation that can be easily adapted to different data types and statistical models.
Cite
Text
Kunze et al. "Practical Shuffle Coding." Neural Information Processing Systems, 2024. doi:10.52202/079017-2672Markdown
[Kunze et al. "Practical Shuffle Coding." Neural Information Processing Systems, 2024.](https://mlanthology.org/neurips/2024/kunze2024neurips-practical/) doi:10.52202/079017-2672BibTeX
@inproceedings{kunze2024neurips-practical,
title = {{Practical Shuffle Coding}},
author = {Kunze, Julius and Severo, Daniel and van de Meent, Jan-Willem and Townsend, James},
booktitle = {Neural Information Processing Systems},
year = {2024},
doi = {10.52202/079017-2672},
url = {https://mlanthology.org/neurips/2024/kunze2024neurips-practical/}
}