Lightnews — Scholar-powered news

Reposted by Sebastian Sanokowski

fses91.bsky.social @fses91.bsky.social · May 22

Happy to introduce 🔥LaM-SLidE🔥!

We show how trajectories of spatial dynamical systems can be modeled in latent space by

--> leveraging IDENTIFIERS.

📚Paper: arxiv.org/abs/2502.12128
💻Code: github.com/ml-jku/LaM-S...
📝Blog: ml-jku.github.io/LaM-SLidE/
1/n

1 8 7

Sebastian Sanokowski @sanokows.bsky.social · Apr 24

11/11 This is joint work with @willberghammer, @haoyu_wang66, @EnnemoserMartin, @HochreiterSepp, and @sebaleh. See you at #ICLR!
[Poster Link](iclr.cc/virtual/202...)
[Paper Link](arxiv.org/abs/2502.08696)
---

Sebastian Sanokowski @sanokows.bsky.social · Apr 24

10/11 🏆 Our method outperforms autoregressive approaches on Ising model benchmarks and opens new avenues for applying diffusion models to a wide range of scientific applications in discrete domains.

1

Sebastian Sanokowski @sanokows.bsky.social · Apr 24

9/11 📊 Due to the mass-covering property of the fKL, it excels at unbiased sampling. Conversely, the rKL is mode-seeking, making it ideal for combinatorial optimization (CO) as it achieves better solution quality with fewer samples.

1 1

Sebastian Sanokowski @sanokows.bsky.social · Apr 24

8/11 💡 𝐒𝐨𝐥𝐮𝐭𝐢𝐨𝐧 2: We address the limitations of the fKL by combining it with Neural Importance Sampling over samples from the diffusion sampler. This allows us to estimate the gradient of the fKL using Monte Carlo integration, making training more memory-efficient.

1 2

Sebastian Sanokowski @sanokows.bsky.social · Apr 24

7/11 An alternative is the forward KL divergence (fKL), where it is well known how to increase memory efficiency by leveraging Monte Carlo integration over diffusion time steps. However, the fKL divergence requires samples from the target distribution!

1

Sebastian Sanokowski @sanokows.bsky.social · Apr 24

6/11 💡 𝐒𝐨𝐥𝐮𝐭𝐢𝐨𝐧 1: We apply the policy gradient theorem to the rKL between joint distributions of the diffusion path. This enables the use of mini-batches over diffusion time steps by leveraging reinforcement learning methods, allowing for memory-efficient training.

1

Sebastian Sanokowski @sanokows.bsky.social · Apr 24

5/11 A commonly used divergence is the reverse KL divergence (rKL), as the expectation of the divergence goes over samples from the generative model. However, naive optimization of this KL divergence requires backpropagating through the whole generative process.

1

Sebastian Sanokowski @sanokows.bsky.social · Apr 24

4/11 🚨 𝐂𝐡𝐚𝐥𝐥𝐞𝐧𝐠𝐞: However, existing diffusion samplers struggle with memory scaling, limiting the number of attainable diffusion steps due to backpropagation through the entire generative process.

1

Sebastian Sanokowski @sanokows.bsky.social · Apr 24

3/11 🔍 𝐃𝐢𝐟𝐟𝐮𝐬𝐢𝐨𝐧 𝐒𝐚𝐦𝐩𝐥𝐞𝐫𝐬 aim to sample from an unnormalized target distribution without access to samples from this distribution. They can be trained by minimizing a divergence between the joint distribution of the forward and reverse diffusion paths.

1

Sebastian Sanokowski @sanokows.bsky.social · Apr 24

2/11 We've developed scalable and memory-efficient training methods for diffusion samplers, achieving state-of-the-art results in combinatorial optimization and unbiased sampling on the Ising model.

1

Sebastian Sanokowski @sanokows.bsky.social · Apr 24

1/11 Excited to present our latest work "Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics" at #ICLR2025 on Fri 25 Apr at 10 am!
#CombinatorialOptimization #StatisticalPhysics #DiffusionModels

1 7 16

Reposted by Sebastian Sanokowski

Amrei Bahr @amreibahr.bsky.social · Jan 10

Starkes Signal!!

Über 60 dt. Hochschulen & Forschungsinstitutionen haben heute ihren Ausstieg bei X bekanntgegeben, s.u. #eXit

X sei nicht mehr vereinbar mit ihren Grundwerten: „Weltoffenheit, wissenschaftliche Integrität, Transparenz und demokratischer Diskurs.“

Liste der Beteiligten hier:

Hochschulen und Forschungsinstitutionen verlassen Plattform X - Gemeinsam für Vielfalt, Freiheit und Wissenschaft

nachrichten.idw-online.de

91 840 3.4K

Reposted by Sebastian Sanokowski

Günter Klambauer @gklambauer.bsky.social · Dec 3

The Machine Learning for Molecules workshop 2024 will take place THIS FRIDAY, December 6.

Tickets for in-person participation are "SOLD" OUT.

We still have a few free tickets for online/virtual participation!

Registration link here: moleculediscovery.github.io/workshop2024/

ML for molecules and materials in the era of LLMs [ML4Molecules]

ELLIS workshop, HYBRID, December 6, 2024

moleculediscovery.github.io

14 19

Sebastian Sanokowski @sanokows.bsky.social · Nov 30

A Pizza Steel or Pizza Stone with max Heat (250 celsius) should do the Job

Sebastian Sanokowski @sanokows.bsky.social · Nov 26

I think it is fine to keep the score, but if all concerns are addressed they should at least justify why they are nevertheless keeping their score.

1

Sebastian Sanokowski @sanokows.bsky.social · Nov 25

Does this mean all Paper at 6 or above should be accepted?

1

Reposted by Sebastian Sanokowski

ICLR Conference @iclr-conf.bsky.social · Nov 25

✍️ Reminder to reviewers: Check author responses to your reviews, and ask follow up questions if needed.

50% of papers have discussion - let’s bring this number up!

1 8 38

Sebastian Sanokowski @sanokows.bsky.social · Nov 23

That is a cool idea!

1 1

Reposted by Sebastian Sanokowski

Marvin Schmitt @marvin-schmitt.com · Nov 22

The ✨ML Internship Feed✨ is here!

@serge.belongie.com and I created this feed to compile internship opportunities in AI, ML, CV, NLP, and related areas.

The feed is rule-based. Please help us improve the rules by sharing feedback 🧡

🔗 Link to the feed: bsky.app/profile/did:...

7 16 63

Sebastian Sanokowski @sanokows.bsky.social · Nov 19

Love seeing the Bluesky community grow!
Just look at the stats—daily activity (likes, posts, and follows) is skyrocketing 📈, with recent peaks such as hitting 3 million daily likes!

Want to explore more about Bluesky’s incredible growth? Check out the live stats page here: bsky.jazco.dev/stats

Atlas - Engagement-Based Social Graph for Bluesky by Jaz (jaz.bsky.social)

bsky.jazco.dev

1 5

Reposted by Sebastian Sanokowski

Jonas Köhler @jonkhler.argmin.xyz · Nov 18

Max Welling (@wellingmax.bsky.social) landed and needs followers! ;)

3 4 32

Sebastian Sanokowski @sanokows.bsky.social · Nov 17

✋

1 1

Sebastian Sanokowski @sanokows.bsky.social · Nov 16

I also would like to join :)

1 2

Reposted by Sebastian Sanokowski

Willie Neiswanger @willieneis.bsky.social · Nov 10

I'm making a list of AI for Science researchers on bluesky — let me know if I missed you / if you'd like to join!

go.bsky.app/AcP9Lix

160 90 250