Georges Le Bellier

Ph.D. student at CNAM on domain adaptation, generative models, and self-supervised learning for remote sensing 🛰
Previously intern @SonyCSL, @Ircam, @Inria

[Georges Le Bellier @_lebellig]

I am excited to share our paper "Detecting Out-Of-Distribution Earth Observation Images with Diffusion Models", accepted at the #CVPR2024 Earthvision workshop! Joint work with @nshaud.

📢 article: https://t.co/wyEwEVn67d

[Georges Le Bellier @_lebellig]

It was mostly a lie. I'm spending most of my social network time in the other place. I'm impressed with how fast the community moved there. My handle is lebellig (same but without the underscore)

[Georges Le Bellier @_lebellig]

It was mostly a lie. I'm spending most of my social network time in the other place. I'm impressed with how fast the community moved there. My handle is lebellig (same but without the underscore)

[Georges Le Bellier @_lebellig]

I made a first scan to find people from here in the other place 🦋, maybe some of them got through: don't hesitate to ping me! (my handle is simply lebellig)
The scrolling experience is so much better there...

[James Thornton @JamesTThorn]

From Diffusion Models to Schrödinger Bridges

- A shame not to see this NeurIPS keynote live, by the incredible @ArnaudDoucet1
- SB naturally extends flow/ bridge matching and diffusion models
- Particularly useful for data to data
- Links to OT
https://t.co/IIpuU0Nw8o

[Yaron Lipman @lipmanya]

A new (and comprehensive) Flow Matching guide and codebase released! Join us tomorrow at 9:30AM @NeurIPSConf for the FM tutorial to hear more...

https://t.co/uaDy00wEw6 https://t.co/ceJlUiTuWO

[Valerio Marsocci @valeriomarsocci]

Are geospatial foundation models really impactful?

🚀🚀🌏

Check it in our new pre-print!

Welcome to **PANGAEA: a global and inclusive benchmark for GFMs**

https://t.co/u2i20cqymO

a short thread 🧵

[Georges Le Bellier @_lebellig]

@t_pelinski are you considering creating a bluesky account?

[Georges Le Bellier @_lebellig]

Do you work on Deep Learning for remote-sensing or Earth Observation (SSL, generative models, foundation models...) and want to move to the other place 🦋?

I've initiated a pack with researchers in the field so we can find each other again! https://t.co/gs2RWBNnMV

[Georges Le Bellier @_lebellig]

Many of you haven't migrated yet 🐦, so don't hesitate to ping me and I'll add you to this starter pack to expand the community!

[Georges Le Bellier @_lebellig]

Do you work on Deep Learning for remote-sensing or Earth Observation (SSL, generative models, foundation models...) and want to move to the other place 🦋?

I've initiated a pack with researchers in the field so we can find each other again! https://t.co/gs2RWBNnMV

[Georges Le Bellier @_lebellig]

I made a first scan to find people from here in the other place 🦋, maybe some of them got through: don't hesitate to ping me! (my handle is simply lebellig)
The scrolling experience is so much better there...

[Georges Le Bellier @_lebellig]

New notebook! To get to grips with discrete flow matching paper, I've put together a little notebook to test it on MNIST. I kept the article's notations so it's easy to follow alongside the article!
🐍 notebookv1: https://t.co/HsHTMjC7wF

[Alain Riou @howariou]

I'll present Stem-JEPA this afternoon at @ISMIRConf! Hope to see you there!

[Georges Le Bellier @_lebellig]

@Charles25110793 Yes, that's why I'm considering using a large number of epochs so I don't get fooled by the "forced convergence". But I think using warmup + reduce lr on plateau is a better way to tune the max number of epochs.

[Georges Le Bellier @_lebellig]

What are the good practices to set up the number of epochs when using a cosine learning rate scheduler? Set it to a large number + early stopping to determine the needed epochs and then rerun the experiment?

[Georges Le Bellier @_lebellig]

I've been on twitter for a long time and really liked it and still appreciate the relevance of so many research posts on this platform. So I will not quit but start cross-posting on 🎆🛤️🏙️ where you can find me @ lebellig

[Georges Le Bellier @_lebellig]

What are the good practices to set up the number of epochs when using a cosine learning rate scheduler? Set it to a large number + early stopping to determine the needed epochs and then rerun the experiment?

[Georges Le Bellier @_lebellig]

Very nice paper! I like the decomposition between the encoder learning the deterministic components+spatial/modality alignment and the flow matching/diffusion learning the stochastic small-scale physics. (+noise scaling to avoid overfitting on the deterministic dynamics)

[Morteza Mardani @MardaniMorteza]

🔍 Superresolution misalignment? Conditional diffusion/flow models falter when low-res and high-res data don’t align.
Meet Stochastic Flow Matching—a framework mapping low-res to latent space, bridging high-res targets effectively.

📝 Paper: https://t.co/LOdfYgpDau

[Georges Le Bellier @_lebellig]

@lostanlen Cool! I'm so sad not to go to his concert in Paris in November

[Günter Klambauer @gklambauer]

Does equivariance matter at scale?

Should a model rather learn equi- and invariances from data or should the architecture have the equiv property? This work provides some insights, and SCALING LAWS for each.

P: https://t.co/dIfSGOvS9N

[Peter Holderrieth @peholderrieth]

New paper out!

We introduce “Generator Matching” (GM), a method to build GenAI models for any data type (incl. multimodal) with any Markov process. GM unifies a range of state-of-the-art models and enables new designs of generative models.

https://t.co/6BTkr3ukYc

(1/5)