Program

Keynote lecture: Reproducibility Challenges: A Focus on Computational Reproducibility with Examples from the Medical Imaging Research Community
Sorina Camarasu-Pop
Biography
Sorina Camarasu-Pop received the Engineering degree in Telecommunications in 2007 and the Ph.D degree (on exploiting heterogeneous distributed systems for Monte-Carlo simulations in the medical field) in 2013, both from the National Institute for Applied Sciences of Lyon (INSA-Lyon, France). Since 2007 she's a CNRS research engineer at the Creatis laboratory in Lyon, France, where she is currently in charge of the VIP platform, which counts more than 1000 registered users. Her activity is focused on optimizing the execution of medical image processing applications on heterogeneous distributed systems. In the last few years, she has been particularly interested in enhancing open and reproducible science through my VIP activities, but also through other projects such as the EU OpenAIRE-Connect, EGI-ACE and EUCAIM projects and the France Life Imaging (FLI) platform, where Ishe is also member of the steering committee of the Information Analysis and Management node. In March 2021, she received the CNRS Médaille de Cristal for having contributed to the excellence of French research. From 2022 to 2024 I was the coordinator of the French ANR ReproVIP (ANR-21-CE45-0024-01) project.
Abstract
This talk begins with a brief overview of the definitions and key concepts of reproducibility, including its multiple layers — from methodological to computational considerations. It then focuses on computational reproducibility, illustrated through studies conducted within the ReproVIP project. Challenges in this area primarily stem from software variability, library dependencies and their evolution over time, as well as numerical instability due to floating-point arithmetic issues (e.g., rounding errors, hardware and compiler optimizations). The talk will show how software, numerical, and hardware variabilities affect linear registration results produced by a neuroimaging application packaged with Docker and Guix (a study presented at ACM REP’24). The presentation concludes with a broader perspective on reproducibility and open questions about addressing its persistent challenges.

Keynote lecture
Simon Tournier
Abstract: TBA