MeLLoC: Lossless Compression with High-order Mechanism Learning

Authors: Xinyue Luo, Jin Cheng, Yu Chen

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through extensive experiments on various scientific datasets, Me LLo C consistently outperforms state-of-the-art lossless compressors while offering compelling trade-offs between compression ratios and computational costs.
Researcher Affiliation Academia Xinyue Luo1 Jin Cheng1,2 Yu Chen2 1School of Mathematical Sciences, Fudan University 2School of Mathematics, Shanghai University of Finance and Economics
Pseudocode Yes Algorithm 1 Compression based on optimizing source term
Open Source Code No The code is not publicly available due to restrictions of an ongoing commercial project. Despite this, the authors are open to academic collaboration, offering an avenue for interested researchers to engage in further communications.
Open Datasets Yes This section presents the experimental results of applying our proposed lossless compression algorithm to the CESM-ATM and Hurricane datasets from the SDRBench[18].
Dataset Splits No The proposed method does not involve neural networks or a training process, and therefore does not specify traditional training/validation splits.
Hardware Specification Yes All tests were conducted on a Mac with M1 Silicon, mac OS 14.1.2, 16GB RAM.
Software Dependencies No The paper does not explicitly provide specific software dependencies with version numbers used for the implementation of Me LLo C.
Experiment Setup Yes We optimize n for coefficients Ci by starting with high precision and gradually reducing it while monitoring reconstruction error and compression ratio.