Gilles Louppe | Lightnews

Light up
your news

About Privacy Terms Help

Gilles Louppe

Gilles Louppe

@glouppe.bsky.social

3.8K followers 850 following 70 posts

AI for Science, deep generative models, inverse problems. Professor of AI and deep learning @universitedeliege.bsky.social. Previously @CERN, @nyuniversity. https://glouppe.github.io

Computer science 62%

Physics 22%

more

Posts Replies Media Videos

Pinned

Gilles Louppe @glouppe.bsky.social · Apr 29

<proud advisor>
Hot off the arXiv! 🦬 "Appa: Bending Weather Dynamics with Latent Diffusion Models for Global Data Assimilation" 🌍 Appa is our novel 1.5B-parameter probabilistic weather model that unifies reanalysis, filtering, and forecasting in a single framework. A thread 🧵

Reposted by Gilles Louppe

Sebastian Raschka (rasbt) @sebastianraschka.com · 2d

Lots of interesting LLM releases last week. My fav was actually Olmo 3 (I love the Olmo series due to their full open-sourceness and transparency).
If you are interested in reading through the architecture details, I coded it from scratch here: github.com/rasbt/LLMs-f...

Gilles Louppe @glouppe.bsky.social · 3d

For some puzzling reason my (existing) Scholar profile cannot be found using Scholar itself, it's been like this for years 🙃 (search engines like Bing do find it however)

Gilles Louppe @glouppe.bsky.social · 3d

... and here I thought the new Scholar Labs would finally be able to search and find my Scholar profile 😥

Reposted by Lisa W. Fazio, Brigitte Nerlich, Ian Miles , and 6 more Lisa W. Fazio, Brigitte Nerlich, Ian Miles, Richard Waite, Pamela K. Gilbert, James Goodwin, Gilles Louppe, Alexandra Heller-Nicholas, David Stott

Colin Dickey @colindickey.com · 4d

Man, everything is so bleak, anyone got a fun fact or little bit of trivia they want to share

Reposted by Gilles Louppe

Allen Institute @alleninstitute.org · 8d

What happens when you combine 10 years of brain data with one of the world’s fastest supercomputers?

A virtual mouse cortex simulation, thanks to a global collaboration.

🧠📈 https://alleninstitute.org/news/one-of-worlds-most-detailed-virtual-brain-simulations-is-changing-how-we-study-the-brain/

Reposted by Gilles Louppe

Tim Kellogg @timkellogg.me · 8d

GDM WeatherNext 2

8x faster than v1, it can compute extreme situations and game out scenarios in one minute flat on a single TPU (as opposed to hours of supercomputer time for traditional algorithms)

will be available in all of Google’s weather apps

blog.google/technology/g...

WeatherNext 2: Our most advanced weather forecasting model

The new AI model delivers more efficient, more accurate and higher-resolution global weather predictions.

Reposted by David M. Markowitz, Gilles Louppe

Michael Deistler @deismic.bsky.social · 12d

I am super happy to share that our project on training biophysical models with Jaxley is now published in Nature Methods: www.nature.com/articles/s41...

Jaxley: differentiable simulation enables large-scale training of detailed biophysical models of neural dynamics - Nature Methods

Jaxley is a versatile platform for biophysical modeling in neuroscience. It allows efficiently simulating large-scale biophysical models on CPUs, GPUs and TPUs. Model parameters can be optimized with ...

Gilles Louppe @glouppe.bsky.social · 13d

Oui c'était bien moi! Je me disais bien aussi que tu me disais qqch :-) au plaisir de se recroiser

Reposted by Gilles Louppe

Gautam Kamath @gautamkamath.com · 14d

It's remarkable how AI generated images have gone from remarkable and visually attractive to repulsive AI slop in just a few years.

Reposted by Gilles Louppe

Kyle Cranmer @kylecranmer.bsky.social · 15d

I am incredibly honored to have received the inaugural AI in Science Research Excellence Prize from the Margot and Tom Pritzker Foundation
dsi.wisc.edu/2025/11/10/d...

Gilles Louppe @glouppe.bsky.social · 15d

‪Once again an interdisciplinary research project gets poor, even condescending this time, evaluations because reviewers make little effort to understand the maths. Science is not only about killing rat models and to see if some random drug worked. Tired of this game.

Reposted by Gilles Louppe

Dorian McILROY @dmcilroy.bsky.social · 17d

"stole Rosalind Franklin's work" has become the new orthodoxy. While she was certainly the victim of sexism from Watson, I think her colleague Wilkins was the real villain. Events 1951-53 well covered in Nature in 2023 www.nature.com/articles/d41...

What Rosalind Franklin truly contributed to the discovery of DNA’s structure

Franklin was no victim in how the DNA double helix was solved. An overlooked letter and an unpublished news article, both written in 1953, reveal that she was an equal player.

Reposted by Gilles Louppe

Kyle Cranmer @kylecranmer.bsky.social · 28d

New paper, with @rkhashmani.me @marielpettee.bsky.social @garrettmerz.bsky.social Hellen Qu. We introduce a framework for generating realistic, highly multimodal datasets with explicitly calculable mutual information. This is helpful for studying self-supervised learning
arxiv.org/abs/2510.21686

Reposted by Gilles Louppe

lebellig @lebellig.bsky.social · 28d

"The Principles of Diffusion Models" by Chieh-Hsin Lai, Yang Song, Dongjun Kim, Yuki Mitsufuji, Stefano Ermon. arxiv.org/abs/2510.21890
It might not be the easiest intro to diffusion models, but this monograph is an amazing deep dive into the math behind them and all the nuances

The Principles of Diffusion Models

This monograph presents the core principles that have guided the development of diffusion models, tracing their origins and showing how diverse formulations arise from shared mathematical ideas. Diffu...

Gilles Louppe @glouppe.bsky.social · 29d

It is only useful when the training data is noisy or incomplete. See eg. arxiv.org/abs/2405.13712 where train diffusion models from sparse images only.

Learning Diffusion Priors from Observations by Expectation Maximization

Diffusion models recently proved to be remarkable priors for Bayesian inverse problems. However, training these models typically requires access to large amounts of clean data, which could prove diffi...

Gilles Louppe @glouppe.bsky.social · 29d

EM algorithm: 1977 vintage, 2025 relevant. New lecture notes on a classic that refuses to age. From fitting a GMM on the Old Faithful data to training modern diffusion models in incomplete data settings, the same simple math applies. 👉 glouppe.github.io/dats0001-fou...

Reposted by Gilles Louppe

Diana Cai @dianarycai.bsky.social · 29d

Fisher meets Feynman! 🤝

We use score matching and a trick from quantum field theory to make a product-of-experts family both expressive and efficient for variational inference.

To appear as a spotlight @ NeurIPS 2025.
#NeurIPS2025 (link below)

Fisher meets Feynman: score-based variational inference with a product of experts

Reposted by Martin Tomko, Gilles Louppe

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · Oct 23

What if we did a single run and declared victory

Three panel thing. In the left panel we use error bars. In the second, we take statistical significance as the biggest number but still have error bars. In LLM science, we just have the biggest number

Reposted by Gilles Louppe

m2lines.bsky.social @m2lines.bsky.social · Oct 15

Excited to share SamudrACE, the first 3D AI ocean–atm–sea-ice #climate emulator! 🚀 Simulates 800 years in 1 day on 1 GPU, ~100× faster than traditional models, straight from your laptop 👩‍💻 Collaboration with @ai2.bsky.social and GFDL, advancing #AIforScience with #DeepLearning.
tinyurl.com/Samudrace

SamudrACE: A fast, accurate, efficient 3D coupled climate AI emulator

A fast digital twin of a state-of-the-art coupled climate model, simulating 800 years in 1 day with 1 GPU. SamudrACE combines two leading…

Reposted by Gilles Louppe

Thomas Fel @thomasfel.bsky.social · Oct 14

🕳️🐇 𝙄𝙣𝙩𝙤 𝙩𝙝𝙚 𝙍𝙖𝙗𝙗𝙞𝙩 𝙃𝙪𝙡𝙡 – 𝙋𝙖𝙧𝙩 𝙄 (𝑃𝑎𝑟𝑡 𝐼𝐼 𝑡𝑜𝑚𝑜𝑟𝑟𝑜𝑤)

𝗔𝗻 𝗶𝗻𝘁𝗲𝗿𝗽𝗿𝗲𝘁𝗮𝗯𝗶𝗹𝗶𝘁𝘆 𝗱𝗲𝗲𝗽 𝗱𝗶𝘃𝗲 𝗶𝗻𝘁𝗼 𝗗𝗜𝗡𝗢𝘃𝟮, one of vision’s most important foundation models.

And today is Part I, buckle up, we're exploring some of its most charming features. :)

Reposted by Peter Thorne, Gilles Louppe

Benjamin Pope @benjaminpope.bsky.social · Oct 14

Thrilled to have two years' of work out, in a pair of papers led by @gradientrider.bsky.social and @maxecharles.bsky.social.

We've built a data-driven calibration of the James Webb Interferometer to near its fundamental limits for high-res imaging - explainer at @aunz.theconversation.com!

How we sharpened the James Webb telescope’s vision from a million kilometres away

The only Australian hardware on board the legendary telescope is starting to fulfil its duties.

theconversation.com

Reposted by Gilles Louppe

Jérémie Kermarrec @jkermarrec.bsky.social · Oct 12

Dans cette interview avec la JASRAC (équivalent japonais de la SACEM), Nobuo Uematsu a donné son avis sur la musique générée par IA. À sa façon, il relaie ce nouvel adage : si personne ne s'est fatigué à l'écrire, je ne me fatiguerai pas à l'écouter.

📃 www.jasrac.or.jp/magazine/int...

Désormais, l’IA générative est capable de composer de la musique. Quelle est votre opinion là-dessus ?
Nobuo Uematsu : Je ne l’ai pas essayée pour le moment, et je pense que je ne le ferai pas. Je suis toujours convaincu qu’il est plus satisfaisant d’avoir créé quelque chose soi-même à la sueur de son front, et que quand on écoute de la musique, une partie du plaisir vient du fait de savoir qui est la personne qui l’a écrite. Avec l’IA, il n’y a pas un tel arrière-plan. Même en termes d’interprétation, la musique tissée par les humains est imprévisible et laissée à la libre appréciation de chacun. Pour moi, c’est précisément cette instabilité, cette effusion qui sont si exaltantes.

Gilles Louppe @glouppe.bsky.social · Oct 10

Bravo Gaël pour cette reconnaissance amplement méritée! Tu dis que la science est un sport d'équipe, et c'est vrai, mais une équipe a besoin d'un leader qui l'inspire. Je suis sûr que l'équipe scikit-learn, passée et présente, est fière de toi et de ce que nous avons construit ensemble. Champagne!

Reposted by Gilles Louppe

Gérôme Andry @gandry.bsky.social · Oct 8

🚀 After more than a year of work — and many great discussions with curious minds & domain experts — we’re excited to announce the public release of 𝐀𝐩𝐩𝐚, our latent diffusion model for global data assimilation!

Check the repo and the complete wiki!
github.com/montefiore-s...

GitHub - montefiore-sail/appa: Code for the publication "Appa: Bending Weather Dynamics with Latent Diffusion Models for Global Data Assimilation".

Code for the publication "Appa: Bending Weather Dynamics with Latent Diffusion Models for Global Data Assimilation". - montefiore-sail/appa

Reposted by Gilles Louppe

Dan Garisto @dangaristo.bsky.social · Oct 1

Had never read Jane Goodall's original 1963 article in Nat Geo on the wild chimpanzees in Tanzania until now. It's a wonderful blend of science and journalism, and well worth your time.

www.nationalgeographic.com/pdf/jane-goo...

David Greybeard, a wild chimpanzee, gets a handout of bananas from author Goodall, who studies the apes under a National Geographic Society grant. This scene near
her camp represents a triumph for Miss Goodall; at first the animals fled if she came
within 500 yards. Knapsack holds camera and notebook. She carries a whistle in her
pocket to summon searchers in case of accident in the rugged Tanganyika hills.

Reposted by Gilles Louppe

Gaël Varoquaux @gaelvaroquaux.bsky.social · Oct 1

Yeah, I think that "AI is the new petrol" is about right...

Reposted by Gilles Louppe

Maxim Raginsky @mraginsky.bsky.social · Sep 27

Rudolf Kalman put it nicely (and provocatively): link.springer.com/chapter/10.1...

Quote from R. Kalman, "Identification from real data": For an objective outsider, much of the historical development
of statistics is a long series of attempts to dodge the inevitable
implications of uncertainty. Whenever the conventional statistical
treatment of a problem gives a unique (certain) answer, as in
maximum likelihood estimation, in least squares, ... , common sense
should tell us that such a miracle is possible only if additional
assumptions (deus ex machina) are imposed on the data which
somehow succeed in neutralizing the intrinsic uncertainty. We
shall use the technical term "prejudice" for such assumptions.
In other words, statistical methodology has been handicapped
because statisticians have become mesmerized by the deep-seated
hope of giving certain answers to problems where the uncertainty
is intrinsic. This is politics, not science. For example, least
squares is very popular because it always gives a unique answer.
But this is exactly where its fatal weakness lies; when we pose
Nature a question, as in identification problems, we must not
phrase that question in such terms that the answer is
predetermined regardless of the nature of the data.

Gilles Louppe @glouppe.bsky.social · Sep 26

How do you type all these long long hyphens? — I keep seeing them everywhere. I learned the UTF-8 code by now but there must be something easier 🤔

Reposted by Gilles Louppe

Evert Nasedkin @evertnasedkin.bsky.social · Sep 26

🔭 It's paper day! Today I'm sharing the latest in a series of papers looking at the weather on other worlds, in this case bringing you the weather report from a nearby T-dwarf, SIMP-0136. 🪐

🧵 to follow...

Reposted by Gilles Louppe

sakanaai.bsky.social @sakanaai.bsky.social · Sep 25

We’re excited to introduce ShinkaEvolve: An open-source framework that evolves programs for scientific discovery with unprecedented sample-efficiency. It leverages LLMs to find state-of-the-art solutions, orders of magnitude faster!

Blog: sakana.ai/shinka-evolve/
Paper: arxiv.org/abs/2509.19349