David Picard

Si seulement. J'aimerais avoir cet optimisme, mais j'ai l'impression que l'idiotie et sa dangerosité sont particulièrement bien réparties dans le monde.

You should at least provide a few links 😅

Ça a toujours été le cas. Quand je faisais mon monitorat thèse, les démarcheurs de Freescale (oui, ça date) venaient nous donner leur microprocesseurs avec des kits dédiés à l'enseignement, avec la condition de faire tout le programme autour de leur tech.

Congrats! 🥳👏 What a journey from your masters day working on metric learning to now! Can't wait to see what you will do next.

I don't know about that second assertion. School looks exactly like that, a fixed dataset that you have to fit, and it's been extremely successful at having us learn centuries of knowledge in a decade or so.
Sure, we learn also using many other means, but that one has been crazy effective.

A reminder that the 70+ year-old research discipline of AI is much bigger than genAI, much bigger than just deep learning, much bigger than Big Tech. AI includes symbolic reasoning, multi-agent systems and statistical machine learning and is used for astrophysics, renewable energy and climate models

ECIR2026 The 48th European Conference on Information Retrieval (ECIR 2026) is Europe's premier forum for cutting-edge research in

📢 The website for ECIR 2026 is live – and so is our first call for papers!

🔗 ecir2026.eu
📍 Delft, The Netherlands
🗓️ March 30 – April 1, 2026 (main conference)
🎓 Tutorials: March 29
🛠 Workshops: April 2

Today's find in the book box at the end of the street. That's a really nice poem.

If we ever build a time machine, we'd have to go back in time to warn Lafayette of what the future is going to look like.

Congrats, you succeeded in having me exhaling a loud curse while reading the votey!

This one was really disgusting. I'll have to clean the bad taste it left in my mouth with the "Brie de Melun" and the "Fourme d'Ambert" I have in my fridge.

This is 😱
Scientific friends, please read these quotes, it is always timely.

I'm doing some verification of image segmentation annotations. One very difficult question that arises is, what is an object? Where should one put a threshold for what makes part of an object or constitutes a different object?

I guess it's subjective, but is anyone aware of relevant literature?

to the use of mathematical morphology as a powerful tool to verify the consistency of the hierarchy as regions are merged.
For example, my colleague Laurent Najman has an entire career worth of publications on the subject, although I'm not sure it will answer your specific question.

This is a question as old as segmentation itself. During my master, I remember already wondering whether the algorithm should put the shirt in the same region as the person wearing it. At the time, long before any learning based method, people "solved" this using segmentation hierarchies, leading

The caveat being that the industry does not fund the same research projects as the public sector.
Or to put it differently: "the cure for tomorrow dies today".

bsky.app/profile/grim...

Pareil, je ne comprends pas pourquoi on utilise Teams.

Which is unfortunately all that PhD students can do with the forum style rebuttal.

There are diminishing returns. Stop trying to improve review quality by excessive extra work.

Remember the irreducible Bayes Error: If the task has inherent subjectivity, it will not help to get better quality measurements.

En tout cas, ce n'est ni bon, ni du pain ! 🤮

But also: I'm not at all against arxiving. Quite the contrary, I'm all for it.
It was a rant about the unfairness of advertising papers under review, which is completely different!

What is a successful promotion of papers? If it's a bunch of likes on a social network, I'm not sure that's super useful.

You know what I feel is the best promotion tool at the moment? Scholar inbox. But that's exactly the opposite of self-advertising by authors. Although the algorithm can be gamed

You can always send a preprint to someone if it's useful. That's how it worked before arxiv and it was as effective as arxiving now.

Everybody misses the 1-page rebuttal.

These lengthy forum style comments are a nightmare: a nightmare for the authors who spend way too much time writing them, a nightmare for the reviewers who spend too much time understanding them, a nightmare for the ACs who will have to summarize all. Stop it!

A three-panel Cyanide & Happiness comic. Panel 1 shows a person telling another they couldn’t access their CV. Panel 2 offers to sell the CV for $39.99 or $10 for 24-hour access. Panel 3 shows the first person entering the Elsevier building

Hey, this is not even a joke #AcademicWriting #AcademicChatter

Press X to doubt.

Est-ce que composer des éléments du corpus peut permettre de générer quelque chose d'atypique ? Par exemple en combinant des éléments qui ne sont pas mis ensemble d'ordinaire ?
Ce n'est pas parce que les modèles génératifs peuvent mémoriser exactement qu'ils doivent nécessairement le faire.

Si possible, j'aimerais bien avoir une explication sur "par définition elle ne pourra rien proposer qui sorte un peu de la norme".

It still happens once every conference! Despite having many awesome colleagues and many many caring friends.

Awesome! That's the way to go! You've been able to do what I've never been able to force myself to do in XX years.

(with an annoyingly large value of XX)

YouTube video by Underscore_ Il a conçu la première IA d’OSINT (terrifiant… et génial)

I had the privilege to be invited to speak about our work "Around the World in 80 Timesteps" at the French Podcast Underscore! If you speak french, i highly recommend it they did a great job with the montage!

If you want to learn more nicolas-dufour.github.io/plonk

www.youtube.com/watch?v=s5oH...