Speaker
Tomasz Procter
(University of Glasgow (GB))
Description
As machine learning becomes increasingly embedded in ATLAS analyses, ensuring the long-term usability of ML-based results poses several challenges. This talk will examine different aspects of ML preservation, including storing and documenting trained models, handling evolving software dependencies, and ensuring accessibility of input features tied to detector conditions. We will discuss the difficulties of rerunning ML-based analyses years after publication and highlight key areas where improvements are needed to support reinterpretation and reproducibility in the future.
Authors
Daniel Thomas Murnane
(Niels Bohr Institute, University of Copenhagen)
Samuel Van Stroud
(UCL)
Tomasz Procter
(University of Glasgow (GB))
