A new TIER2 preprint, titled "Reproducibility in Machine Learning-based Research: Overview, Barriers and Drivers", has been accepted for publication in the AI Magazine. Co-authored by TIER2's Dominik Kowald and Tony Ross-Hellauer alongside their colleagues Harald Semmelrock, Simone Kopeinik, Dieter Theiler, Armin Haberl, and Stefan Thalmann.

The study examines the lack of transparency, data or code, poor adherence to standards, and the sensitivity of machine learning training conditions, which are the reasons many papers are not even reproducible in principle. Where they are, reproducibility experiments have found low degrees of similarity with original results.

The publication presents a new perspective on the key procedural and technical barriers to reproducibility at different levels (methods, code, data, and experiments). It also identifies key drivers that can help address these issues and maps them to specific barriers, offering practical strategies for researchers to mitigate reproducibility in their work. Additionally, it highlights areas where further research is needed and calls for a more serious commitment from the machine learning community to tackle these challenges.