Calvin and Hobbes! Making me feel bad about my job as a ML researcher.

This semester I have been given the chance to teach a course on "Deep Learning for Time Series" at @ipparis.bsky.social If you're interested in the topics, have a look at rtavenar.github.io/x-dl4ts/ Repo has Typst code for the slides and Python code for the labs. Feedback is very much welcome.

This year I'm teaching a new course on generative models for visual content (images, video, 3D, etc). It's mostly me rambling about recent papers, design choices I like/hate. The slides of the first lectures are here: davidpicard.github.io/teaching/ Use right arrow to navigate past the blank page.

Next time you want to use a vibe coded compile think about benchmarking the resulting binary. ccc on SQLite is up to 150k slower on some SQL queries and the benchmark took 2 hours vs 10sec for GCC compiled harshanu.space/en/tech/ccc-...

So how does it work? Meet Miniature (mini + IA, get it?), a complete AI agent in ~225 lines of Python. No frameworks, no abstractions. Code: github.com/glouppe/miniature

We are recruiting four positions connected to Machine Learning, Statistical Learning, and AI for Science in the Applied Mathematics department at École polytechnique. Join our vibrant community at IP Paris and Hi! Paris IA center. List below🧵 tinyurl.com/3jpw9t26

Early Arthropods xkcd.com/3199/

This reddit post made me laugh

Apparently the NeurIPS chairs decided to reopen the camera-ready submission to ensure "that the proceedings reflect the highest scientific quality.” @neuripsconf.bsky.social, is that true? So the solution to academic misconduct is to give the authors a chance to cover their tracks?

The good news is that we did not receive any ICLR rejects today. The bad is that we might start writing ICML papers on Monday😱.

NeurIPS reviewers typically review 6+ papers, each with 100 or so citations. The reviewers can't possibly check that every citation is real. But why don't we have an automated way to do this checking (and desk-rejecting if fake citations found) *before* the papers are sent to reviewers?

Accurate and thorough representation of prior and related work is one of the cornerstones of good research. It is shocking to me that so many published NeurIPS papers, even from top institutions, have fabricated references. I recommend reading the original report: gptzero.me/news/neurips/

This is a unique opportunity that should not be missed

Open position: Monge Assistant Professor (Tenure track) on graph learning and/or LLMs, in the applied math dpt. at École polytechnique. Reduced teaching load, many funding opportunities, good students and academic freedom. Contact me if interested. Deadline March 23rd tinyurl.com/MongeAssista...

Doing a PhD is - at heart - one long discussion with your mentor. The discussion changes over time - with unexpected turns and ups & downs - but through it all is a pair of people discussing a topic endlessly to make sense of it. PhD students: choose someone you like to talk to!

Telling your students about research before the ImageNet moment

I missed this post but it is pure gold. www.colincornaby.me/2025/08/in-t...

Openreview opened the door to continuous and major revisions that nobody has time to check properly. I think that we should come back to short one pdf page replies to reviews. It would mean having decisions quicker so that we actually have time to work on papers before resubmitting them.

really annoying. my student just made this joke: ICLR= I Can Locate Reviewers 😅

OpenReview was breached. The names of authors, reviewers, ACs, etc, for all past and current conferences were visible for a time, making nothing anonymous anymore. These data have been released for this year's ICLR, but I fear it's also the case for the past 10 years of conferences.

AI researchers loose privacy. I guess this is karma for what we have done with the data from everyone 😉. ICLR PC need to freeze all open review forum changes right now because the process is tainted for all future edits

Sonia and I will present this work this afternoon at the NeuripS@Paris poster session. Come see us if you are around.

I am happy to share that I will be at NeurIPS in San Diego to present our paper with @rflamary.bsky.social and Bertrand Thirion on optimal transport plan prediction between graphs. If you are around come say hi! Paper: arxiv.org/abs/2506.12025 Poster #3703 Friday 5 December 4:30 - 7:30 pm

I was worried for a sec but since it's sent from the US this is totally a legitimate scientific publisher and not a predatory journal at all.

Une raison de plus de vous abonner à la chaîne YouTube du colloque #GRETSI : "Transport optimal, de Monge à l’apprentissage profond", conférence plénière de Julie Delon au colloque #GRETSI2025 www.youtube.com/watch?v=ujYS...

Going to buy this one for my reviews and reply to reviewers. tampographe.com/products/bul...

Kyle Kingsbury is not a journalist. He is not an op-ed writer. He is a computer safety researcher. And he has written one of the most compelling, comprehensive accounts of the ongoing hell in Chicago that you could possibly imagine. In under 1600 words. aphyr.com/posts/397-i-...

This Nature retrospective is quite interesting. To me, the only solution to the credit assignment problem is obvious: stop believing a single person is responsible for every big discovery. It's an artifact of our monkey brain requiring a face for storage, not the reality of how knowledge progresses.

Yesterday @tgnassou.bsky.social successfully defended his PhD thesis on domain adaptation of signals and in particular EEG. Huge congrats to him for all his work and his wonderful slides. It was a pleasure to be his advisor with @agramfort.bsky.social and I can't wait to see what he will do next!

The Amphora of Great Intelligence (AGI) Part 2 #webcomic #krita #miniFantasyTheater

What if we did a single run and declared victory

Competitor enters a major AI competition (RNA folding) GPU poor so can't train an AI Builds a "classic" eng pipeline instead. (90s tech) Wins and beat everyone using DL 💀 Their winning "hybrid" model had an AI in it. Their original one did not and had a higher score So they won despite the AI 😂

Probably the best paper title of my career. To read with the Indiana jones soundtrack. And yes we solved and differentiated quite well 80 millions (small) Fused Gromov Wasserstein problems per epoch using a neural network on GPU.

Our latest paper “The Quest for the GRAph Level autoEncoder (GRALE)” was accepted at NeurIPS 2025! arxiv.org/abs/2505.22109 🏆 GRALE 🏆 can encode and decode graphs into and from a shared Euclidean space. Training such a model should require solving the graph matching problem but...

A speech about what drives me, how science and open source are bitter victories, unable to make improve the world if society does not embrace them for the better: gael-varoquaux.info/personnal/a-...

It's a great discussion because we all have different interpretation of what this means and what the consequences are. Myself, I have absolutely no doubt that scaling works. If you have all the videos in the world and are able to train a model that can recall and merge any of them, then for sure...

Figure 1. Happy ML researcher and open source developer presenting his toolbox SKADA at PyData Paris. Congrats @tgnassou.bsky.social the presentation was awesome!

I am happy to share that our paper "Unsupervised Learning for Optimal Transport plan prediction between unbalanced graphs" was accepted at Neurips 2025 ! 🥳 Huge thanks to my co-authors @rflamary.bsky.social and Bertrand Thirion ! arxiv.org/abs/2506.12025 (1/5)

Did not know that all authors that submit 3 ICLR of more must be reviewers with 6 (or more?) papers. Seems fair to review 2 papers per submission but after the full NeurIPS summer (and rejection) I would have appreciated a reduced load to do them better (I say that as an ICML 2025 top reviewer).

I missed it last month! Very happy to present POT next week at PyData Paris.

📢 Talk Announcement "Optimal Transport in Python: A Practical Introduction with POT", by Rémi Flamary. 📜 Talk info: pretalx.com/pydata-paris-2025/talk/FNCKXU 📅 Schedule: pydata.org/paris2025/schedule 🎟 Tickets: pydata.org/paris2025/tickets

SKADA is a beautiful software for Domain Adaptation in python with many shallow and deep methods implemented, it is 100% compatible with @scikit-learn.org models and pipelines and with @pytorch.org for deep learning methods.

📢 Talk Announcement "Tackling Domain Shift with SKADA: A Hands-On Guide to Domain Adaptation", by Théo Gnassounou and Antoine Collas. 📜 Talk info: pretalx.com/pydata-paris-2025/talk/MEPTCE 📅 Schedule: pydata.org/paris2025/schedule 🎟 Tickets: pydata.org/paris2025/tickets

Our article "Large-scale semi-discrete optimal transport with distributed Voronoi diagrams" (with Nicolas Ray, Quentin Merigot and Hugo Leclerc) just got accepted by Journal of Computational Physics. Pushing the limits of Optimal Transport by several orders of magnitude !

I'm thrilled to announce that my #ERCStG project **Optinfinite : Efficient infinite-dimensional optimization over measures** has been accepted. Thank you @erc.europa.eu ! Many thanks also to @crestumr.bsky.social @ipparis.bsky.social for they support, as well as to my collaborators and friends.

To: Reviewer 2 My name is Inigo Montoya You killed my paper Prepare to die

Thanks for raising this issue. As someone on the authors' and reviewers' side, this feels like a violation of the reviewing process. One might wonder, “Why do we put so much time and effort into the peer-review process if decisions are ultimately made arbitrarily?”.

I wrote about the tragic death of Adam Raine and the venal negligence of "AI Safety." www.argmin.net/p/the-banal-...

Let's push for the obvious solution: Dear @neuripsconf.bsky.social ! Allow authors to present accepted papers at EurIPS instead of NeurIPS rather than just additionally. Likely, at least 500 papers would move to Copenhagen, problem solved.