Journée du Deep Learning pour la Science – Après-midi du 8 juin 2026

Value of the Information & Strength of the Argument

The video provides valuable insights into the current state of AI applications in various scientific fields, particularly in meteorology and biology. The speakers present concrete examples and results, demonstrating the practical benefits and limitations of deep learning approaches. The argumentation is generally solid, with speakers acknowledging challenges and uncertainties, such as the difficulty in predicting extreme weather events and the need for model interpretability. However, the talks are introductory and do not delve deeply into technical details, which may limit their value for experts. The inclusion of multiple perspectives from different disciplines enriches the content, but the depth of each presentation is constrained by the short time allocated.

Scientific Rigor, Source Quality, Title Accuracy

The video maintains a high level of scientific rigor, with speakers from reputable institutions presenting ongoing research. The sources cited are primarily the speakers’ own work and well-known datasets like ERA5, but specific references are not always provided in the talk. The title accurately reflects the content, as it is a recording of the afternoon sessions of the JDLS. The description includes a link to the FIDLE website, which serves as a source for further information. Overall, the content is reliable and well-presented, though the lack of detailed citations within the talks may be a minor weakness.

The title accurately reflects the content: it is a recording of the afternoon sessions of the Deep Learning for Science day, covering multiple talks on AI applications in various scientific fields.

8/10

The video features expert speakers from recognized institutions (Météo France, Sorbonne Université, CNRS) presenting ongoing research and results. The content is well-structured, with clear explanations of methodologies and limitations. However, as a recorded conference, it lacks peer review and some claims are presented without detailed evidence.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction and welcome remarks by the host, mentioning updates on FIDLE and upcoming content.
• 05:00Talk by Laure Raynaud (Météo France) on AI in weather forecasting, starting with an overview of traditional forecasting steps.
• 20:00Discussion of hybrid and fully data-driven models, including the AIFS model and its architecture.
• 40:00Presentation of results showing AI models outperforming physical models in temperature prediction, but struggling with extreme events like the Valencia floods.
• 1:00:00Talk by Alessandra Carbone on AI for protein function prediction, covering structure and systems biology.
• 1:20:00Break and transition to the next talk.
• 1:30:00Talk by Barthélémy Jobert on digital art history and AI, focusing on Delacroix.
• 1:50:00Talk by Olivier Wong-Hee-Kam on digital infrastructure, sovereignty, and sustainability.
• 2:10:00Award ceremony for best presentations and posters.
• 2:20:00Closing remarks by Jean-Luc Parouty and end of the event.

• FIDLE website — Mentioned in the video description as the official website for the FIDLE training program.

• FIDLE website — Official source for the training program and related resources.