The above project got the 2021 R&D 100 Award. The project does compression of the data with relatively high compression ratio without loosing significant information. It is used in projects related to cosmology, fluid dynamics and so on.
More information on this is here: https://indico.fnal.gov/event/57188/contributions/254705/attachments/162007/213971/ATLAS-CMS.pdf (Probably only available if you have FNAL credentials unfortunately).
https://www.alcf.anl.gov/events/accelerate-python-loops-intel-ai-analytics-toolkit
UC Irvine maintains an archive of data for ML training. Data on over 600 different topics:
Check out their GPU-FPGA Training lessons
https://tac-hep.org/training-modules/uw-gpu-fpga