Découvertes de nébuleuses avec l'intelligence artificielles. Nicolas Outters et Dominique Daniel

Value of the Information & Strength of the Argument

The presentation provides valuable insights into the practical application of AI in astronomy, specifically for the detection of planetary nebulae. The speakers share their hands-on experience, including the difficulties encountered with CNNs and the decision to switch to Swin Transformers. They also discuss the importance of data augmentation and the trade-off between precision and recall. The argumentation is coherent and grounded in their direct involvement, though it lacks quantitative results or performance metrics. The value lies in the detailed description of the workflow, from AI training to human verification, and the emphasis on the collaborative aspect between amateurs and professionals.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is moderate: the speakers reference known catalogs (HASH, Aladin) and the work of Quentin Parker’s team, but they do not provide specific citations or publications. The title accurately reflects the content, and the talk is well-structured. The lack of formal references is compensated by the practical experience shared. The audience questions are addressed, but the discussion remains at a descriptive level rather than a rigorous scientific analysis.

The title accurately reflects the content: the speakers discuss their project of discovering planetary nebulae using AI, with Nicolas Outters and Dominique Daniel as the main presenters.

7/10

The presentation is based on the speakers' direct involvement in the project and references to known astronomical databases and catalogs. However, no formal citations or peer-reviewed sources are provided, and the technical details are presented at a high level.

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

• 02:00Introduction and genesis of the project: discussion about AI and planetary nebulae.
• 05:00Presentation of the AIP association and its 450 members.
• 10:00Context: 5,000 known planetary nebulae, estimated 30,000 in the sky.
• 15:00Challenges: competition from professional teams, need for small and faint objects.
• 20:00Use of CNNs and their limitations, shift to Swin Transformers.
• 25:00Partnership with Central Digital Lab and student involvement.
• 30:00Data augmentation techniques to expand the training set.
• 35:00Importance of human verification and follow-up imaging.
• 40:00Tools used: Aladin, HASH catalog, and validation process.
• 45:00Conclusion and future prospects.

• Aladin Sky Atlas — Mentioned as a tool for visual verification and comparison with catalogs.
• HASH Planetary Nebula Database — Referenced as a catalog of known planetary nebulae for validation.

• Quentin Parker's team publications — Mentioned as a professional team using similar AI techniques, with hundreds of discoveries.