Pinned
A new (and comprehensive) Flow Matching guide and codebase released! Join us tomorrow at 9:30AM @NeurIPSConf for the FM tutorial to hear more...
arxiv.org/abs/2412.06264 github.com/facebookresearโฆ
GIF
Yaron Lipman
336 posts
Yaron Lipman
@lipmanya
Research scientist @AIatMeta (FAIR), prev @WeizmannScience. Interested in generative models and deep learning of irregular/geometric data.
Israel
Joined August 2014
- Our **Flow Matching Tutorial** from #NeurIPS2024 is now publicly available: neurips.cc/virtual/2024/tโฆ @helibenhamu @RickyTQChen
- **Flow Matching** (#ICLR2023 spotlight) offers a simple simulation-free method for training flow-based generative models, generalizing and improving upon diffusion models in training speed, sampling efficiency, and generation quality. @RickyTQChen @helibenhamu @mnick @lematt1991
GIF - New paper: Introducing Moser Flows (MFs), a new class of continuous normalizing flows (CNFs) on manifolds based on divergences of neural nets. First generative modeling results on general curved surfaces! with Noam Rozen @adityagrover_ @mnick arxiv.org/abs/2108.08052 1/7
- Excited to share that our paper "Moser Flow: Divergence-based Generative Modeling on Manifolds" won an Outstanding Paper Award at NeurIPS 2021!! blog.neurips.cc/2021/11/30/annโฆ Noam Rozen @adityagrover_ @mnick See thread for a short summary ๐New paper: Introducing Moser Flows (MFs), a new class of continuous normalizing flows (CNFs) on manifolds based on divergences of neural nets. First generative modeling results on general curved surfaces! with Noam Rozen @adityagrover_ @mnick arxiv.org/abs/2108.08052 1/7
- New paper: Universal Differentiable Renderer (UDR) - learning 3D shapes from 2D images under a wide range of reflectance and lighting models. arxiv.org/abs/2003.09852 @YarivLior @matanatzmon
00:00 - New paper: Noticing a surprising incarnation of implicit regularization of neural networks in the geometric setting, and harnessing its power for high fidelity shape representation. arxiv.org/abs/2002.10099 @AmosGropp @YarivLior @HaimNiv @matanatzmon
- "Generator Matching" is a significant generalization of Flow Matching. It demonstrates that the "matching" principle applies to a wide range of Markov processes, opening up a vast, unexplored design space for scalable generative models.New paper out! We introduce โGenerator Matchingโ (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. arxiv.org/abs/2410.20587 (1/5)
- ๐ฃ A new #ICML2023 paper investigates the Kinetic Energy of Gaussian Probability Paths which are key in training diffusion/flow models. A surprising takeaway: In high dimensions *linear* paths (Cond-OT) are Kinetic Optimal! Led by @shaulneta w/ @RickyTQChen @lematt1991 @mnick
- For those interested in *implicit neural representations* my talk from #cvpr20 workshop on Deep Learning Foundations of Geometric Shape Modeling and Reconstruction is here: youtu.be/rUd6qiSNwHs featuring work by @matanatzmon @YarivLior @AmosGropp and others.
- Mosaic-SDF is a **simple 3D shape representation**, based on a set of local grids that is: (1) param efficient, (2) easy to compute for a given shape, and (3) โtensorialโ. Therefore, it is suitable for scalable 3D generative models.
00:00
00:03Thrilled to share Mosaic-SDF (M-SDF), a simple 3D representation suitable for 3D generative models! Check out more results of text-to-3D generations here - lioryariv.github.io/msdf - Designing both expressive and efficient equivariant (or invariant) neural architectures can be done by replacing the costly group averaging with an *input dependent* subset, called a Frame.
Excited to share โFrame Averaging for Invariant and Equivariant Network Designโ: a general framework for building computationally efficient and maximally expressive equivariant neural architectures. Accepted as an *Oral presentation* to #ICLR2022! arxiv.org/abs/2110.03336 - My presentation at the ELLIS workshop on Geometric and Relational Deep Learning: youtu.be/fveyx5zKReo via @YouTube
- ๐ฃ We are looking for a PhD intern interested to work with us on generative models in FAIR EMEA (specifically, Paris)!Yaron @lipmanya and I are hiring a PhD intern for FAIR EMEA! If you're interested in fundamental research on generative modeling and related topics, feel free to reach out: {rtqichen,ylipman}@meta.com. *The position is in Paris.* Dates are flexible. metacareers.com/jobs/140947795โฆ
