Lightnews — Scholar-powered news

Reposted by Mathurin Massias

Tony S.F. @tonysf.bsky.social · 19d

My paper on Generalized Gradient Norm Clipping & Non-Euclidean (L0, L1)-Smoothness (together with collaborators from EPFL) was accepted as an oral at NeurIPS! We extend the theory for our Scion algorithm to include gradient clipping. Read about it here arxiv.org/abs/2506.01913

3 14

Mathurin Massias @mathurinmassias.bsky.social · 19d

merci David !

Mathurin Massias @mathurinmassias.bsky.social · 19d

merci !

Mathurin Massias @mathurinmassias.bsky.social · 19d

Our work on the generalization of Flow Matching got an oral at Neurips !

Go see @quentinbertrand.bsky.social present it there :)

Mathurin Massias @mathurinmassias.bsky.social · Jun 18

New paper on the generalization of Flow Matching www.arxiv.org/abs/2506.03719

🤯 Why does flow matching generalize? Did you know that the flow matching target you're trying to learn *can only generate training points*?

w @quentinbertrand.bsky.social @annegnx.bsky.social @remiemonet.bsky.social 👇👇👇

3 3 25

Reposted by Mathurin Massias

Pierre Marion @pierremarion.bsky.social · 22d

🔥 Excited to announce the Workshop on the Principles of Generative Models at @euripsconf.bsky.social (the European conference parallel to NeurIPS 2025)
🇩🇰 Dec 6–7, Copenhagen
📝 Deadline for contributions: Oct 17
🔗 Website: sites.google.com/view/prigm-e...

PriGM@EurIPS2025

Workshop summary Machine learning theory has long focused on classical supervised learning settings, where a model is trained on input–label pairs drawn from a well-defined data distribution, with the...

sites.google.com

2 5

Mathurin Massias @mathurinmassias.bsky.social · 29d

Félicitations Anna !!

1 1

Mathurin Massias @mathurinmassias.bsky.social · Sep 4

Oui, tout sera en anglais !

1

Mathurin Massias @mathurinmassias.bsky.social · Sep 4

Oui !

1

Mathurin Massias @mathurinmassias.bsky.social · Sep 4

Oui... c'est un compromis avec le fait d'avoir suffisamment de créneaux et de temps de discussions aux posters. Tu peux éventuellement arriver un peu après le début, et sinon ça devrait être accessible à distance 🤞

1 1

Mathurin Massias @mathurinmassias.bsky.social · Sep 4

One-day workshop on Diffusion models and Flow matching, October 24th at @ensdelyon.bsky.social

Registration and call for contributions (short talk and poster) are open at
gdr-iasis.cnrs.fr/reunions/mod...

Modèles génératifs : diffusion, flow matching et leurs applications - GdR IASIS

Les modèles génératifs ont connu de récentes avancées spectaculaires, au point que leurs dernières versions sont désormais capables de produire des images et du texte synthétiques presque indiscernabl...

gdr-iasis.cnrs.fr

2 7 10

Mathurin Massias @mathurinmassias.bsky.social · Jun 27

Super elegant approach !

Rémi Flamary @rflamary.bsky.social · Jun 27

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...

3

Mathurin Massias @mathurinmassias.bsky.social · Jun 27

on second thoughts I'm not sure I understood. In the classical FM loss you do have to learn this derivative no ? The loss is :

1

Mathurin Massias @mathurinmassias.bsky.social · Jun 27

I was thinking of this:

1

Mathurin Massias @mathurinmassias.bsky.social · Jun 26

I was thinking of the linear interpolant yes ; I haven't seen papers where other are used

1

Mathurin Massias @mathurinmassias.bsky.social · Jun 26

it could even be velocity matching, and this time you do learn match the *conditional* velocities

1 3

Mathurin Massias @mathurinmassias.bsky.social · Jun 26

Thanks for the kind words

2

Mathurin Massias @mathurinmassias.bsky.social · Jun 18

Then why does flow matching generalize?? Because it fails!

The inductive bias of the neural network prevents from perfectly learning u* and overfitting.

In particular neural networks fail to learn the velocity field for two particular time values.

See the paper for a finer analysis 😀

3 5

Mathurin Massias @mathurinmassias.bsky.social · Jun 18

We propose to regress directly against the optimal (deterministic) u* and show that it never degrades the performance
On the opposite, removing target stochasticity helps generalizing faster.

1

Mathurin Massias @mathurinmassias.bsky.social · Jun 18

Yet flow matching generates new samples!

An hypothesis to explain this paradox is target stochasticity: FM targets the conditional velocity field ie only a stochastic approximation of the full velocity field u*

*We refute this hypothesis*: very early, the approximation almost equals u*

1 3

Mathurin Massias @mathurinmassias.bsky.social · Jun 18

New paper on the generalization of Flow Matching www.arxiv.org/abs/2506.03719

🤯 Why does flow matching generalize? Did you know that the flow matching target you're trying to learn *can only generate training points*?

w @quentinbertrand.bsky.social @annegnx.bsky.social @remiemonet.bsky.social 👇👇👇

2 17 55

Mathurin Massias @mathurinmassias.bsky.social · Apr 24

On Saturday Anne will also present some very, very cool work on how to leverage Flow Matching models to obtain sota Plug and Play methods:

PnP-Flow: Plug-and-Play Image Restoration with Flow Matching, poster #150 in poster session 6, Saturday at 3 pm

arxiv.org/abs/2410.02423

PnP-Flow: Plug-and-Play Image Restoration with Flow Matching

In this paper, we introduce Plug-and-Play (PnP) Flow Matching, an algorithm for solving imaging inverse problems. PnP methods leverage the strength of pre-trained denoisers, often deep neural networks...

arxiv.org

1 1 6

Mathurin Massias @mathurinmassias.bsky.social · Apr 24

It was received quite enthusiastically here so time to share it again!!!

Our #ICLR2025 blog post on Flow Matching was published yesterday : iclr-blogposts.github.io/2025/blog/co...

My PhD student @annegnx.bsky.social will present it tomorrow in ICLR, 👉poster session 4, 3 pm, #549 in Hall 3/2B 👈

A Visual Dive into Conditional Flow Matching | ICLR Blogposts 2025

Conditional flow matching (CFM) was introduced by three simultaneous papers at ICLR 2023, through different approaches (conditional matching, rectifying flows and stochastic interpolants). <br/> The m...

iclr-blogposts.github.io

1 6 11

Reposted by Mathurin Massias

Samuel Vaiter @samuelvaiter.com · Apr 15

The proximal operator generalizes projection in convex optimization. It converts minimisers to fixed points. It is at the core of nonsmooth splitting methods and was first introduced by Jean-Jacques Moreau in 1965. www.numdam.org/article/BSMF...

3 21

Reposted by Mathurin Massias

Julian Tachella @tachellajulian.bsky.social · Apr 14

🚢🚢 deepinv v0.3.0 is here, with many new features! 🚢 🚢

Our passionate team of contributors keeps shipping more exciting tools!

Deepinverse (deepinv.github.io) is a library for solving imaging inverse problems with deep learning.

Redirecting to https://deepinv.github.io/deepinv/

deepinv.github.io

1 4 11

Mathurin Massias @mathurinmassias.bsky.social · Apr 14

I had a blast giving a summer school on generative models at AI Hub Senegal, in particular flow matching, with @quentinbertrand.bsky.social and @remiemonet.bsky.social

Our material is publicly available !!! github.com/QB3/SenHubIA...

ensdelyon.bsky.social

1 3 17