Deutsch Intern
Compression Methods for Robustness and Transferability (COMFORT)

News and Events

On February 25 and 26, 2026, we held our second annual meeting for our COMFORT project at the German Electron Synchrotron (DESY) in Hamburg. In several presentations and discussion rounds, the project progress was presented and the next steps were discussed. A key insight supporting the COMFORT hypothesis was that compressed models can generalize at least as well as dense models, while requiring significantly less training effort and sometimes even being more robust to attacks.

more

From 05-06 February 2025 we had the kick-off meeting for our project COMFORT here at the Julius-Maximilians-Universität Würzburg. We had two days of exciting discussions on how the compression of machine learning models interacts with robustness and transferability.

more

The development and use of artificial intelligence consumes a lot of resources. The new BMBF-funded research project COMFORT aims to change this. The mathematician Leon Bungert from Würzburg is responsible for the project. (Article only in German)

more