Reproduce “Foundation Model for Lossy Compression of Spatiotemporal Scientific Data” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xiao Li, Jaemoon Lee, Anand Rangarajan, Sanjay Ranka
Researchers at the University of Florida developed a foundation model for lossy compression of spatiotemporal scientific data by integrating a variational autoencoder with a hyper-prior structure and a super-resolution module. This approach yielded up to 4x higher compression ratios compared to traditional methods like SZ3 and ZFP on unseen scientific datasets while maintaining comparable reconstruction quality.
Key insights
- Integrating a hyper-prior VAE with an SR module effectively learns generalizable representations for diverse spatiotemporal scientific data.
- A hybrid approach combining 2D and 3D convolutions efficiently captures complex spatiotemporal correlations in scientific data while maintaining computational feasibility.
- Sophisticated super-resolution techniques within the decoder significantly boost compression efficiency and reconstruction quality compared to simpler upsampling methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.