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.

A university and an academic publisher walk into a bar. The publisher orders a pint, sells it back to the university, asks the barman to pay the bill.

I wrote an op-ed on the world-class STEM research ecosystem in the United States, and how this ecosystem is now under attack on multiple fronts by the current administration: newsletter.ofthebrave.org/p/im-an-awar...

🤖 Dang, we are seeing increasing evidence of this: The use of generative AI negatively affecting critical thinking. www.bloomberg.com/news/article... Paging @emilymbender.bsky.social, one of the few people who started warning the public about this years ago...

By now it's just too much information (reviews, rebuttal back-and-forth, final reviewer justification, final author remark). On average, for each paper I'm reviewing there is a discussion of ~10 pages. It's almost impossible to keep track of everything.

From: www.argmin.net/p/the-negron...

I came here to kick ass AND chew bubble gum And I’m all out of gum Therefore it is impossible to kick ass AND chew bubble gum I should have come to kick ass XOR chew bubble gum

Why #NeurIPS PC? Why? We already have days of reading to do including reviews, responses, responses to responses and confidential comments to AC. Do you really think a "Final Remark" that tells me to accept the paper will help? Let me just do my AC job during august while everyone is on vacation!

Hum I have some scores disappearing on openreview on my #NeurIPS papers. Do you have the same? Maybe it means the scaore has been edited? If this is to hide the score until final decision, this seems like a bad idea (we know how to do data imputation in the community)

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!

As a #NeurIPS AC I did a little stat. Paper reviews+responses in my batch contain 515k characters and 75k words. This is without including papers themselves or authors/reviewers discussions. So please be nice to your AC, we are doing our best. Also I miss the one page PDF response of old ICML.

This prompt injection for GEMINI CLI is just fascinating. It's crazy that because of LLM, hackers can now do social engineering without a human in the loop. But I guess giving execution rights to an LLM is already a big humain failure anyways. arstechnica.com/security/202...

SKADA-Bench : Benchmarking Unsupervised Domain Adaptation Methods with Realistic Validation On Diverse Modalities, has been published published in TMLR today 🚀. It was a huge team effort to design (and publish) an open source fully reproducible DA benchmark 🧵1/n. openreview.net/forum?id=k9F...

I love the #eurips initiative! But to live up to its potential, it should be accepted an official #neurips conference location (similar to Mexico City) and not just an addon! Then we would save C02, rather than adding to it!!! What can we do to achieve this?

This is awful! We need to treat our scientists better if we want to do good science. This means consistent funding for tenure that allows time for risky but novel discoveries and not only short-time projects and infinite post-docs. www.science.org/content/arti...

✨Thrilled to see EurIPS launch — the first officially endorsed European NeurIPS presentation venue! 👀 But NeurIPS now requires at least one author to attend in San Diego or Mexico (and not just virtually as before). This is detrimental to many. Why not allow presenting at EurIPS or online? 1/4

Edwige Cyffer presenting her work on the interaction between decentralization and privacy after receiving the 2025 best PhD prize from the French machine learning scientific society (SFFAM) at CAP 2025.

@rflamary.bsky.social 's keynote at CAp'25 was a blast! A short summary in the linked post below...

The most important aspect when facing data shift is the type of shift present in the data. I will give below a few examples of shifts and some existing methods to compensate for it.🧵1/6

Very honored to have given the keynote at the CAP 2025 conference in Dijon this morning. I talked about Domain adaptation, DA benchmark and applications of DA to sleep stating. Mostly work from the PhD of @tgnassou.bsky.social For those interested here are the slides remi.flamary.com/pres/OTDA_Ke...

Time to take a vacation!

Distributional Reduction paper with H. Van Assel, @ncourty.bsky.social, T. Vayer , C. Vincent-Cuaz, and @pfrossard.bsky.social is accepted at TMLR. We show that both dimensionality reduction and clustering can be seen as minimizing an optimal transport loss 🧵1/5. openreview.net/forum?id=cll...

Having a drink after the first day of the skada code sprint. Thanks to all the participants tons of nice things are coming. scikit-adaptation.github.io

Our work on simulation-based inference at the LHC is featured in @arstechnica.com! Well done @aishikghosh.bsky.social @sandesarajay.bsky.social 🙌 arstechnica.com/science/2025...

I gave a POT sticker to Aymeric Dieuleveut and he found the perfect place for it: his Paris public transit card.

📣 New preprint 📣 **Differentiable Generalized Sliced Wasserstein Plans** w/ L. Chapel @rtavenar.bsky.social We propose a Generalized Sliced Wasserstein method that provides an approximated transport plan and which admits a differentiable approximation. arxiv.org/abs/2505.22049 1/5

Just found out about this weird videon and I love it. Automated subtitles are decent. Watch till then end ! youtu.be/1gSnrlIp9XI?...

💡 To make it easier for everyone to apply these techniques, we built Skada: a simple, Python-based library for Domain Adaptation. Skada team organizes a coding sprint :(@rflamary.bsky.social , @antoinecollas.bsky.social , @ambroiseodt.bsky.social) 📍 When: June 24–25 📍 Where: Inria Saclay

Skada Sprint Alert: Contribute to Domain Adaptation in Python 📖 Machine learning models often fail when the data distribution changes between training and testing. That’s where Domain Adaptation comes in — helping models stay reliable across domains.

While overleaf is down, let us take a deep breath and imagine the paper fits in 9 pages. And not think about those typos we cannot fix!

A paper I wrote in 2019 has just been published. Academia is a game of patience. If you wait long enough, Reviewer 2 will die first.

I have cleaned a bit my lecture notes on Optimal Transport for Machine Learners arxiv.org/abs/2505.06589

PhD Timeline xkcd.com/3081

The fast inverse square root trick from Quake III is an algorithm to quickly approximate 1/√x, crucial for 3D graphics (normalisation). It uses bit-level manipulation and Newton's method for refinement. web.archive.org/web/20170729...

People think that data scientists are replacing them with AI. The scrub team is fighting back by replacing data scientists and all their practical knowledge with Python classes.

Aussi aujourd'hui je retombe par hasard sur ce banger:

We have been reworking the Quickstart guide of POT to show multiple examples of OT with the unified API that facilitates access to OT value/plan/potentials. It allows to select regularization/unbalancedness/lowrank/Gaussian OT with just a few parameters. pythonot.github.io/master/auto_...

Open source is draining. One's todo-list is open to world. People seldom realize the cost of what they get for free. Unpleasant comments do happen. Cost of maintenance is not understood. opensource.com/article/17/2...

A friend included this anti-acknowledgement section on her PhD thesis. She also added the proposition: “Systematic bullying and undermining of girls and women in STEM starts early on and is the reason why they do not stay in science and related fields.” Absurd we still need to go through this 🧪👩‍🔬

One of the major limit of modern ML and CV is huge energy and data requirement. This is an impressive result that shows that one can do very well with less. A refreshing reality check that smart always wins over brute force! They need synthetic caption from existing models though.