19.11.2024

On 19 November 2024, the Appl-FM team met for a project meeting at the Einstein Center Digital Future in Berlin. Felix Biessmann presented on realistic error models for tabular data: errors are important for regularization, data augmentation, and privacy, yet little research has addressed realistic error models. With better modelling of realistic errors, he outlined, we can improve training, make robustness evaluations more realistic, and better understand how errors are generated and where they come from.

The work was published as Towards Realistic Error Models for Tabular Data by Philipp Jung, Sebastian Jäger, Nicholas Chandler, and Felix Biessmann, and continues in the Realistic Errors in Tabular Data project.