Excited to share that our paper on sprite-based image decomposition is accepted at TMLR! 🎉 Sprite-based models are highly interpretable but struggle to scale to complex, multi-object images. 
 1/3

Understand-based research vs BOLD numbers. Submit your work to our ECCV workshop on Empirical Theory: empiricaltheory.github.io @davidpicard.eurosky.social @a-lucic.bsky.social m.loog, h.hung, h.pinson.

Qui aurait pu prédire ? « Testée lors des JO de Paris, l’expérimentation a ensuite été prolongée par la loi olympique mais n’était prévue que jusqu’en 2027. » Caméras algorithmiques : l’élargissement de l’expérimentation jusqu’à la fin de 2030 validée par les députés www.lemonde.fr/socie...

Reviewer 2 refusing to enjoy a good paper 😆

The deadline is in a few days (July 13, Sydney time)! #USyd usyd.wd105.myworkdayjobs.com/en-US/USYD_E...

We appreciate the love ❤️ But limited engagement and few followers on Bsky makes some (not us!) question whether it's worth it Be sure to follow (and share with your ML friends!) if you want to see more updates from ICML!

Excited to share our #ECCV2026 paper introducing Whareformer, which learns to track what is where in long egocentric videos. Proud to collaborate with Jacob Chalk and @sinhasaptarshi.bsky.social and the inspiring, humbling forces that are @dlarlus.bsky.social and @dimadamen.bsky.social

The Bayeux Tapestry has arrived in London for a 1y display in the British Museum. It's such a gorgeous piece of art and history. I'm so happy for all the people who will be able to enjoy it there! www.lemonde.fr/culture/arti...

🌟Discussions avec Jean-Jacques Hublin : « Confidences de Paléoanthropologue » 🌟Dans le cadre de notre exposition gratuite « #Préhistoire : entre utopie et réalité », nous avons interrogé le Pr Hublin ! ⏰ L'exposition reste à découvrir du jusqu’au au 19 juillet 2026 au Collège de France !

the thing is, this shouldn't be sobering, this shouldn't be _at all_ surprising, scientists have been screaming from the rooftops that this was on the way for longer than most people on planet earth have been alive, and yet we chose to pretend things were fine and to burn more and more carbon

🏹 Job alert: 2 PhDs & 1 PostDoc in Robust Agentic AI at JKU Linz 📍 Linz 🇦🇹 ⏰ 15 August 🔗 https://bit.ly/4vr051r

I haven't realised that @thomasbrox.bsky.social is on Bluesky! Hi Thomas! 👋 Smash the follow button!

#3DV2027 #CallForPapers 3dvconf.github.io/2027/call-fo...

NeurIPS not having folks sign up for emergency reviewing and just expecting ACs to track people down is just baffling. I was lucky to find someone to agree, but found myself very much missing the process in CV where folks sign up to serve as emergency reviewers and that info is in OpenReview.

Super urgent, I'm desperately trying to find two emergency reviewers for Neurips on a super interesting paper that study how LLM alignements evaluation depends on many different geopolitical factors. Seriously, I've been trying for a week, no one answered.

so true

To me this is not new and not surprising: (Applied) ML always had a very strong experimental component and finding out what trained models can do is akin to studying biological systems. This has only become more important when data, models and tasks have become complex. Nothing strange about this.

We all like to play offended, but what if it's the same trend in real jobs? Either we have to agree that the role of high ed is not to learn how to do a job, or we have to agree that AI is a capability boon that is to transform jobs. It is not a xor btw, at least one of these propositions is true.

A Brown professor gave his students a take-home midterm exam. After suspecting many cheated using AI, he made the final in-person. The orange dots are the midterm scores and the gray dots are the final scores. I applaud S22 for honesty because I would've cheated.

Oh, do I have one?

📄Le HCC publie son rapport annuel 2026 : Dangers climatiques : la France face à ses responsabilités. 🔗Lien du rapport : www.hautconseilclimat.fr/publications...

Waiting for a doom rewrite in utf-8 transliteration

Vous avez mal écrit : "Le jour où elle a été reconnue coupable pour la deuxième fois d'avoir détourné des millions d'euros d'argent public mais ou les médias lui ont déroulé le tapis rouge pour qu'elle puisse chouiner"

I have just resigned from the board of "Statistics and Computing", along with 17 other Associate Editors. This was a difficult decision: this great journal has published many tremendous articles under the leadership of EiC Ajay Jasra, and before him David Hand, Gilles Celeux, and Mark Girolami.

C’est ça qui devrait faire la une de tous les journaux. Un jeune qui avait la vie devant lui est mort pour servir ses concitoyens. Alors qu’on sait qu’il faut sortir des énergies fossiles, on s’étripe bien au frais sur des solutions miracles qui seront vite obsolètes. Et des gens meurent. Colère.🖤

C'est vrai que c'est du délire. Elle s'est faite débarquer parce qu'elle préfère que les services de l'Etat utilisent des outils gratuits et efficaces plutôt que payants et bordéliques. Et qu'elle refuse d'appauvrir encore l'Etat au profit d'actionnaires de boites privées.

Summer Witchcraft Sources and bonus timelapse: www.peppercarrot.com/en/miniFanta... #webcomic #krita #miniFantasyTheater

I had Fable build another thing I always wanted, a full procedural fantasy kingdom generator with economics, trade routes, population growth, wars, lineages, and occasional dragons. First, I worked with it on a plan, then it made it happen. You can play it here: annals-kingdom.netlify.app

It's a funny jab, but it's not true. You can read very carefully all the papers you can, it will not prevent you to copy/paste a buggy bibtex from Google scholar. The only thing preventing bad references is a well curated, manually maintained, handwritten bibtex that you keep from papers to papers.

There's a demo of the model btw: huggingface.co/spaces/nicol...

I'm sadly not at ICML, but @arrijitghosh.bsky.social and @lucasdegeorge.bsky.social are presenting MIRO right now! Happening right now at poster board 2508!

Une rupture d'égalité devant la loi qui n'a pas sa place dans un état de droit...

"On a un aperçu du climat que l’on va vivre dans 10 ou 20 ans: on pourra connaître des canicules qui s’étalent sur un mois, prévient le climatologue @davifaranda.bsky.social. On doit s’interroger sur les politiques que l’on met en place pour s’adapter à cette nouvelle normalité" 357/x

Call for emergency reviewers #NeurIPS2026 I'm looking for two emergency reviewers: - Multimodal (text + vision) benchmark - AI for Science (text-focused) benchmark If you have experience in either domains, send me a DM including your OpenReview profile and Google Scholar links

Venice https://www.wikiart.org/en/gustave-moreau/venice

"I want to open the black box that is LLMs and see what they are thinking!" Do you though? bsky.app/profile/isol...

Gave a few bucks to help having infrastructures hosted and operated in the EU

The biggest challenge of the ICML 2026 review process was the number of submissions. It doubled in size! The acceptance rate remained roughly stable.

The PC Chairs are presenting the welcome remarks. #ICML2026 has over 20,000 in-person attendees!

20000! Oh my !#&@% god! Twenty thousands... For ICML. This makes no sense I don't have words. I mean, it's consistent with the number of submissions, but still. I always viewed ICML as kind of niche. In 2019, I remember ICML was just before CVPR and had less than half the attendance. What happened?

P-hacking is a long-standing problem in science, and LLMs make it worse: as tools for annotation or LLM-as-a-judge, they allow tuning parameters until a desired result appears. We propose a way to mitigate this problem, and conduct a preregistered study of its effectiveness! arxiv.org/abs/2606.27687

What even is theory? Nice discussion here: empiricaltheory.github.io/synopsis/

⚠️ Le nombre de vagues de chaleur s’accélère depuis les années 2000. C’est l’une des conséquences du changement climatique.  🌍 Sur les 53 vagues de chaleur en France depuis 1947, plus de la moitié ont eu lieu après...2010 ! 🔎 Pour en savoir plus : meteofrance.com/le-changemen...

La cheffe de file de l'extrême droite avait conditionné sa candidature à l'absence de condamnation au port d'un bracelet. Elle doit s'exprimer au journal de 20h de TF1.

Excellent Keynote by @akorba.bsky.social at Cap-Rfiap in Montpellier, France, on machine learning and distances between distributions.

We just released POT 0.9.7 with many new OT solvers including lazy and sparse exact solvers, HD Gaussians, sliced OT plans, Spectral-Grassmann OT, batch exact OT with Proximal point, BSP-OT, unbalanced 1D and semidiscrete. This is a big one! Thanks to all contributors! github.com/PythonOT/POT...

Random post that is somehow timely

Good news: We can tell you how much the globe will warm if we continue to burn fossil fuels and emit CO₂ at an alarming rate. Bad news: We’re terrible at telling you how bad the impacts will be, and they’re consistently worse than we predicted.