Lossless data compression by large models

doi:10.1038/s42256-025-01033-7

	Lossless data compression by large models
	Li, Ziguang 1,2,3; Huang, Chao 4,5; Wang, Xuliang 1,4,6; Hu, Haibo 4,5; Wyeth, Cole 6; Bu, Dongbo 1,4; Yu, Quan 2; Gao, Wen 2; Liu, Xingwu 1,7; Li, Ming 1,2,3,6
	2025-05-01
发表期刊	NATURE MACHINE INTELLIGENCE
卷号	7 期号:5 页码:794-799
摘要	Data compression is a fundamental technology that enables efficient storage and transmission of information. However, traditional compression methods are approaching their theoretical limits after 80 years of research and development. At the same time, large artificial intelligence models have emerged, which, trained on vast amounts of data, are able to 'understand' various semantics. Intuitively, semantics conveys the meaning of data concisely, so large models hold the potential to revolutionize compression technology. Here we present LMCompress, a new method that leverages large models to compress data. LMCompress shatters all previous lossless compression records on four media types: text, images, video and audio. It halves the compression rates of JPEG-XL for images, FLAC for audio and H.264 for video, and it achieves nearly one-third of the compression rates of zpaq for text. Our results demonstrate that the better a model understands the data, the more effectively it can compress it, suggesting a deep connection between understanding and compression.
DOI	10.1038/s42256-025-01033-7
收录类别	SCI
语种	英语
资助项目	Canadian Network for Research and Innovation in Machining Technology, Natural Sciences and Engineering Research Council of Canada (NSERC Canadian Network for Research and Innovation in Machining Technology)[2022YFA1304603] ; National Key R&D Program of China ; Proteomic Navigator of the Human Body Project[OGP0046506] ; Canada's NSERC ; Canada Research Chair Program[62072433] ; Canada Research Chair Program[62088102] ; Canada Research Chair Program[62025101] ; National Natural Science Foundation of China[2024Z119] ; Kechuang Yongjiang 2035 key technology breakthrough plan of Zhejiang Ningbo
WOS研究方向	Computer Science
WOS类目	Computer Science, Artificial Intelligence ; Computer Science, Interdisciplinary Applications
WOS记录号	WOS:001479636100001
出版者	NATURE PORTFOLIO
引用统计
文献类型	期刊论文
条目标识符	http://119.78.100.204/handle/2XEOYT63/40629
专题	中国科学院计算技术研究所期刊论文_英文
通讯作者	Liu, Xingwu; Li, Ming
作者单位	1.Cent China Inst Artificial Intelligence, Zhengzhou, Peoples R China 2.Peng Cheng Lab, Shenzhen, Peoples R China 3.Shanghai Inst Math & Interdisciplinary Sci, Shanghai, Peoples R China 4.Chinese Acad Sci, Inst Comp Technol, Beijing, Peoples R China 5.Ningbo Inst Artificial Intelligence Ind, Ningbo, Peoples R China 6.Univ Waterloo, Sch Comp Sci, Waterloo, ON, Canada 7.Dalian Univ Technol, Sch Math Sci, Dalian, Peoples R China
推荐引用方式 GB/T 7714	Li, Ziguang,Huang, Chao,Wang, Xuliang,et al. Lossless data compression by large models[J]. NATURE MACHINE INTELLIGENCE,2025,7(5):794-799.
APA	Li, Ziguang.,Huang, Chao.,Wang, Xuliang.,Hu, Haibo.,Wyeth, Cole.,...&Li, Ming.(2025).Lossless data compression by large models.NATURE MACHINE INTELLIGENCE,7(5),794-799.
MLA	Li, Ziguang,et al."Lossless data compression by large models".NATURE MACHINE INTELLIGENCE 7.5(2025):794-799.