At Ital-IA in Naples 2024, the research “Dictionary Learning for Data Compression within a Digital Twin Framework,” developed thanks to EuroCC Italy

On May 29, 2024, at the Ital-IA in Naples, the national conference dedicated to Artificial Intelligence and organized by CINI (Consorzio Interuniversitario Nazionale per l’Informatica), an important research will be presented, developed by Laura Cavalli of CINECA HPC Department, developed thanks to the “Working on the field” internship of the EuroCC training program.

The paper “Dictionary Learning for Data Compression within a Digital Twin Framework” presents the engineering of a workflow “DL4DT” for optimizing data transfer within a Digital Twin, drastically reducing communication time between the edge and the Cloud. The developed tool is based on the Dictionary Learning compression method. By transferring a significantly smaller amount of data (up to 80% reduction), this method achieves AI algorithm training with the same level of accuracy, reducing update times for deploying new models to be used in production on the edge.

The presented workflow is capable of operating efficiently on a wide range of datasets, including images and time series, and is particularly well-suited for implementation on devices with limited computational resources, making it usable on the edge.

The applicability of the workflow extends beyond data compression; it can also be used as a pre-processing technique for noise reduction, enhancing data quality for subsequent training. The library will soon be released on the EuroCC Italy GitHub page and will be available in open-source mode for Italian companies to optimize their digital twins.