Announcements

Very happy to announce our latest paper “The Cost of Reproducibility in Artificial Intelligence” has been published in Patterns! You can read the paper here (Its open access!)

Highlights

• We analyze 918 AI publications between 2022 and 2024 for their reproducibility
• Papers published at ML conferences are overall easiest to reproduce
• Award-winning papers are not easier to reproduce than “regular” papers
• We find a dark pattern of 6.72% of URLs to code repositories being broken

Big thank you to my co-authors Holger and Odd Erik (my supervisors) as well as my coworkers who sacrificed a substantial amount of their time to help me build this papers’ data set!

Our new paper, “Automated Reproducibility has a Problem Statement Problem”, has been accepted at the Reproducible Artificial Intelligence workshop @ AAAI 2026! Read our paper, on why automating reproducibility is currently problematic and how the problem statement can be formulated, here; ArXiv

Our code is now available in the PyTorch Geometric package, see their documentation website for details: Read the Docs

Our paper “Edge-Based Graph Component Pooling” won the AstraZeneca Healthcare & Bio Award at MLG workshop! We’re happy and honoured with this recognition of our work.

Our paper “Edge-Based Graph Component Pooling” was accepted at the 21st International Workshop on Mining and Learning with Graphs! Thanks again to Bram Renting and Jan van Rijn at Leiden University for the great collaboration.