🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

If data lives on a manifold, how do we design meaningful interpolations between marginals? We present Metric Flow Matching (MFM)… Peter Potaptchik Teo Reu Leo Zhang Alex Tong Michael Bronstein Joey Bose Francesco Di Giovanni 🔗Dive in here: arxiv.org/abs/2405.14780 🧵 (1/12)

Congratulations to Mahdi, Idan, Karolina and Roi for receiving an oral presentation at ICML for work that significantly deepens our understanding between memorization and generalization in stochastic convex optimization.

With Peter Potaptchik and George Deligiannidis we show the first realistic bound on the iteration complexity of diffusion models! Our work explains why sampling from ImageNet needs only ~100 (intrinsic dim) steps instead of ~150k (extrinsic dim). arxiv.org/abs/2410.09046

Michael Churchill Peter Potaptchik Teo Reu Leo Zhang Alex Tong Michael Bronstein Joey Bose Francesco Di Giovanni github.com/kksniak/metric…

At #NeurIPS and interested in Flow Matching? Come and find us on Thursday presenting MFM - a generalization of FM with probability paths supported on the data manifold 🗺️ Poster: #2411 East Exhibit Hall A-C ⏲️Thu 12 Dec 11 am-2pm neurips.cc/virtual/2024/p… Reach out to chat!

How annealing helps overcoming multimodality? In our ICLR 2025 paper openreview.net/forum?id=P6IVI… and preprint arxiv.org/abs/2502.04575, we established the first complexity bound for annealed sampling and normalizing constant (⇔free energy) estimation under weak assumptions on target!

In our new paper (accepted at ICLR!), we propose the first framework for constructing equivariant diffusion models via symmetrisation This allows us to ensure E(3)-equivariance with just highly scalable standard architectures such as Diffusion Transformers, instead of EGNNs, for