Open Science in the Age of AI: Reproducible Research Project
M2 MVA · Semester 2 · 5 ECTS
Contents
- Introduction and overview
- What the module asks
- How to prepare
- Deliverables and assessment
- Official sources
Introduction and overview
| Module | Open Science in the Age of AI: Reproducible Research Project |
| Official title | Science ouverte à l’ère de l’IA – Projet de recherche reproductible / Open science in the age of AI – Reproductible research project |
| Instructors | M. COLOM BARCO (ENS Paris-Saclay, per master-mva.com, 2026–2027) |
| Period | Semester 2 · period 2 (04/01/2027 – 02/04/2027) |
| Credits and volume | 5 ECTS · 21 h (tutorials 9 h, labs 12 h) per the Université Paris-Saclay programme table; master-mva.com announces 8 sessions of 3 h, mainly practical |
| Label | Track RR · no Learning / Modelling label on master-mva.com |
| Language | not displayed |
| Assessment | Each group delivers a reproducible article as a complete IPOL pre-print (article, source code and online demo) and an evaluation report on another group’s project. Final score: the reproducible project 80 %, the intermediate practical work 20 %. |
| Prerequisites | not displayed; presented as the practical follow-up of “Fundamentals of reproducible research and free software”, although the two courses are independent |
This is a framing page, not a course: the module is a tutored project. Students read a scientific article in detail, implement the method, publish a complete scientific report with its source code, and build an online demonstrator in the IPOL journal system.
Each year the MVA lecturers propose articles that deserve a reproducible edition; projects may be co-supervised by researchers of Centre Borelli or other research centres.
What the module asks
- Write a reproducible article: introduction with the main ideas, method and architecture, data, training and inference for neural methods, experiments, limitations and conclusion.
- Match the code to the pseudo-code, at the right level of mathematical and algorithmic detail.
- Review another group’s work, which puts every student in both the author and the reviewer role.
The short introductory sessions cover writing (structure, style, analysis of good and bad articles), tools (online execution and dissemination platforms, peer-reviewed journals, git and GitHub or GitLab, LaTeX and Overleaf, agile team organisation), and coding practice (seeded pseudo-random generators, choice of libraries, assertions of invariants, granularity of functions, documentation, and what reproducibility means for a neural network: algorithm, architecture, weights and data).
How to prepare
- Follow Fundamentals of Reproducible Research and Free Software in semester 1, or read its page: version control, environments, testing, licences and the IPOL model.
- Pick an article in a field you master: many course pages of these notes end with a paper-study project that can grow into an IPOL edition — for example Image Denoising, Sub-Pixel Image Processing or Detection Theory.
- Plan the demo early: an online demonstrator constrains the run time and the input formats of the code.
Deliverables and assessment
- A complete IPOL pre-print per group: the article, the source code and an online demo.
- An evaluation report on the project of another group.
- Weights: the final reproducible project 80 %, the intermediate practical work 20 %.
These are the official rules from master-mva.com; this page adds no assessment of its own.
Official sources
- master-mva.com, course page “Science ouverte à l’ère de l’IA – Projet de recherche reproductible”, 2026–2027: https://www.master-mva.com/cours/projet-de-recherche-reproductible/.
- master-mva.com, Track Recherche reproductible: https://www.master-mva.com/formation-mva/track-recherche-reproductible/.
- IPOL Journal — Image Processing On Line: https://www.ipol.im/.