Lightnews — Scholar-powered news

Arnaud Doucet

@arnauddoucet.bsky.social

890 followers 220 following 10 posts

Senior Staff Research Scientist @Google DeepMind, previously Stats Prof @Oxford Uni - interested in Computational Statistics, Generative Modeling, Monte Carlo methods, Optimal Transport.

Posts Media Videos Starter Packs

Pinned

Arnaud Doucet @arnauddoucet.bsky.social · Dec 15

The slides of my NeurIPS lecture "From Diffusion Models to Schrödinger Bridges - Generative Modeling meets Optimal Transport" can be found here
drive.google.com/file/d/1eLa3...

BreimanLectureNeurIPS2024_Doucet.pdf

drive.google.com

9 67 330

Reposted by Arnaud Doucet

majhas.bsky.social @majhas.bsky.social · Jun 10

(1/n)🚨Train a model solving DFT for any geometry with almost no training data
Introducing Self-Refining Training for Amortized DFT: a variational method that predicts ground-state solutions across geometries and generates its own training data!
📜 arxiv.org/abs/2506.01225
💻 github.com/majhas/self-...

1 4 12

Reposted by Arnaud Doucet

λ³🎲 @cubiclogic.bsky.social · Jun 20

Shunichi Amari has been awarded the 40th (2025) Kyoto Prize in recognition of his pioneering research in the fields of artificial neural networks, machine learning, and information geometry

www.riken.jp/pr/news/2025...

甘利俊一栄誉研究員が「京都賞」を受賞

甘利俊一栄誉研究員（本務：帝京大学先端総合研究機構特任教授）は、人工ニューラルネットワーク、機械学習、情報幾何学分野での先駆的な研究が評価され、第40回（2025）京都賞（先端技術部門　受賞対象分野：情報科学）を受賞しました。

www.riken.jp

2 12 35

Reposted by Arnaud Doucet

LOGML Summer School @logml.bsky.social · Mar 11

🌟Applications open- LOGML 2025🌟

👥Mentor-led projects, expert talks, tutorials, socials, and a networking night
✍️Application form: logml.ai
🔬Projects: www.logml.ai/projects.html
📅Apply by 6th April 2025
✉️Questions? [email protected]

#MachineLearning #SummerSchool #LOGML #Geometry

LOGML 2025

London Geometry and Machine Learning Summer School, July 7-11 2025

logml.ai

2 9 20

Arnaud Doucet @arnauddoucet.bsky.social · Mar 9

Just write a short informal email. If the person needs a long-winded polite email to answer, then perhaps you don't want to have to interact with them.

1 1

Reposted by Arnaud Doucet

Kirill Neklyudov @k-neklyudov.bsky.social · Mar 6

SuperDiff goes super big!
- Spotlight at #ICLR2025!🥳
- Stable Diffusion XL pipeline on HuggingFace huggingface.co/superdiff/su... made by Viktor Ohanesian
- New results for molecules in the camera-ready arxiv.org/abs/2412.17762
Let's celebrate with a prompt guessing game in the thread👇

1 4 14

Reposted by Arnaud Doucet

Masud Husain @masudhusain.bsky.social · Mar 6

Why academia is sleepwalking into self-destruction. My editorial @brain1878.bsky.social If you agree with the sentiments please repost. It's important for all our sakes to stop the madness
academic.oup.com/brain/articl...

51 310 540

Reposted by Arnaud Doucet

Grant Rotskoff @grant.rotskoff.cc · Mar 4

Excited to see our paper “Computing Nonequilibrium Responses with Score-Shifted Stochastic Differential Equations” in Physical Review Letters this morning as an Editor’s Suggestion! We uses ideas from generative modeling to unravel a rather technical problem. 🧵 journals.aps.org/prl/abstract...

1 2 10

Arnaud Doucet @arnauddoucet.bsky.social · Mar 5

Great intro to PAC-Bayes bounds. Highly recommended!

Pierre Alquier @pierrealquier.bsky.social · Mar 5

I already advertised for this document when I posted it on arXiv, and later when it was published.

This week, with the agreement of the publisher, I uploaded the published version on arXiv.

Less typos, more references and additional sections including PAC-Bayes Bernstein.

arxiv.org/abs/2110.11216

2 12

Arnaud Doucet @arnauddoucet.bsky.social · Feb 8

Well you can do it but we don't have any proof. We actually also ran alpha-DSBM with zero-variance noise, i.e. so really an "alpha-rectified flow": experimentally it does "work" but we have no proof of convergence for the procedure.

Arnaud Doucet @arnauddoucet.bsky.social · Feb 8

Yes the trajectories are not quite smooth as they correspond to a Brownian bridge and, as the variance of the reference Brownian motion of your SB goes to zero, you get back to the deterministic and straight paths of OT.

1 1 4

Reposted by Arnaud Doucet

Arthur Gretton @arthurgretton.bsky.social · Feb 5

Better diffusions with scoring rules!

Fewer, larger denoising steps using distributional losses; learn the posterior distribution of clean samples given the noisy versions.

arxiv.org/pdf/2502.02483

@vdebortoli.bsky.social Galashov Guntupalli Zhou @sirbayes.bsky.social @arnauddoucet.bsky.social

arxiv.org

1 8 29

Arnaud Doucet @arnauddoucet.bsky.social · Jan 16

A standard ML approach for parameter estimation in latent variable models is to maximize the expectation of the logarithm of an importance sampling estimate of the intractable likelihood. We provide consistency/efficiency results for the resulting estimate: arxiv.org/abs/2501.08477

On the Asymptotics of Importance Weighted Variational Inference

For complex latent variable models, the likelihood function is not available in closed form. In this context, a popular method to perform parameter estimation is Importance Weighted Variational Infere...

arxiv.org

2 30

Arnaud Doucet @arnauddoucet.bsky.social · Jan 10

Speculative sampling accelerates inference in LLMs by drafting future tokens which are verified in parallel. With @vdebortoli.bsky.social , A. Galashov & @arthurgretton.bsky.social , we extend this approach to (continuous-space) diffusion models: arxiv.org/abs/2501.05370

10 45

Arnaud Doucet @arnauddoucet.bsky.social · Jan 8

I personally read at least a couple of hours per day. It is not particularly focused and I might "waste" time but I just enjoy it.

1 1 14

Arnaud Doucet @arnauddoucet.bsky.social · Dec 27

Very nice paper indeed. I like it.

Reposted by Arnaud Doucet

Joey Bose @joeybose.bsky.social · Dec 18

🔊 Super excited to announce the first ever Frontiers of Probabilistic Inference: Learning meets Sampling workshop at #ICLR2025 @iclr-conf.bsky.social!

🔗 website: sites.google.com/view/fpiwork...

🔥 Call for papers: sites.google.com/view/fpiwork...

more details in thread below👇 🧵

2 19 84

Reposted by Arnaud Doucet

lebellig @lebellig.bsky.social · Dec 17

Schrödinger Bridge Flow for Unpaired Data Translation (by @vdebortoli.bsky.social et al.)

It will take me some time to digest this article fully, but it's important to follow the authors' advice and read the appendices, as the examples are helpful and well-illustrated.

📄 arxiv.org/abs/2409.09347

10 22

Arnaud Doucet @arnauddoucet.bsky.social · Dec 15

The slides of my NeurIPS lecture "From Diffusion Models to Schrödinger Bridges - Generative Modeling meets Optimal Transport" can be found here
drive.google.com/file/d/1eLa3...

BreimanLectureNeurIPS2024_Doucet.pdf

drive.google.com

9 67 330

Reposted by Arnaud Doucet

Alex Thiery @alexxthiery.bsky.social · Dec 15

One #postdoc position is still available at the National University of Singapore (NUS) to work on sampling, high-dimensional data-assimilation, and diffusion/flow models. Applications are open until the end of January. Details:

alexxthiery.github.io/jobs/2024_di...

18 41

Arnaud Doucet @arnauddoucet.bsky.social · Dec 14

I couldn't speak for the following 3 days :-)

Reposted by Arnaud Doucet

Gabriel Peyré @gabrielpeyre.bsky.social · Dec 4

I have updated my course notes on Optimal Transport with a new Chapter 9 on Wasserstein flows. It includes 3 illustrative applications: training a 2-layer MLP, deep transformers, and flow-matching generative models. You can access it here: mathematical-tours.github.io/book-sources...

2 19 100

Reposted by Arnaud Doucet

Gergely Neu @neu-rips.bsky.social · Dec 4

exciting new work by my truly brilliant postdoc Eugenio Clerico on the optimality of coin-betting strategies for mean estimation!

for fans of: mean estimation, online learning with log loss, optimal portfolios, hypothesis testing with E-values, etc.

dig in:
arxiv.org/abs/2412.02640

On the optimality of coin-betting for mean estimation

Confidence sequences are sequences of confidence sets that adapt to incoming data while maintaining validity. Recent advances have introduced an algorithmic formulation for constructing some of the ti...

arxiv.org

2 8 41