Peter Holderrieth (@peholderrieth) / X

Peter Holderrieth

91 posts

Peter Holderrieth

@peholderrieth

CS PhD student at @MIT • Generative Modeling and AI4Science • Prev: Stats/Neuro @OxfordUni• Math at @UniBonn • Former: @AIatMeta

Joined July 2022

Pinned
Peter Holderrieth
@peholderrieth
Apr 22
We release Diamond Maps💎 unlocking accurate and efficient guidance for diffusion models. Our experiments show that our methods scale incredibly well. Excited to see what people will build with this! Accurate guidance has been a notoriously hard problem, but in this work, we’re
00:00
59K
Peter Holderrieth
@peholderrieth
Mar 3, 2025
Our MIT class “6.S184: Introduction to Flow Matching and Diffusion Models” is now available on YouTube! We teach state-of-the-art generative AI algorithms for images, videos, proteins, etc. together with the mathematical tools to understand them. diffusion.csail.mit.edu (1/4)
257K
Peter Holderrieth
@peholderrieth
Oct 30, 2024
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. arxiv.org/abs/2410.20587 (1/5)
91K
Peter Holderrieth
@peholderrieth
Feb 18, 2025
New paper out! We introduce “LEAPS”, a neural sampling algorithm for discrete distributions via continuous-time Markov chains (“discrete diffusion”). We introduce a novel importance sampling scheme and novel symmetries built into neural networks. arxiv.org/pdf/2502.10843 (1/4)
17K
Peter Holderrieth
@peholderrieth
Mar 3, 2025
Replying to @peholderrieth
We include: 1. Lecture notes deriving flow matching and diffusion models (diffusion.csail.mit.edu/docs/lecture-n…) 2. Lecture videos (youtube.com/watch?v=GCoP2w…) 3. Labs that guide you to code up your own model from scratch (diffusion.csail.mit.edu). (2/4)
7K
Peter Holderrieth
@peholderrieth
Apr 25, 2025
Come to our oral presentation on Generator Matching at ICLR 2025 tomorrow (Saturday). Learn about a generative model that works for any data type and Markov process! Oral: 3:30pm (Peridot 202-203, session 6E) Poster: 10am-12:30pm #172 (Hall 3 + Hall 2B) arxiv.org/abs/2410.20587
6.2K
Peter Holderrieth
@peholderrieth
Mar 3, 2025
Replying to @peholderrieth
Flow and diffusion models are mathematically demanding subjects - which is why many lectures restrict themselves to teaching high level intuition. Here, we give a mathematically rigorous and self-contained introduction yet aimed at beginners in AI. We hope you will like it!
4.4K
Peter Holderrieth
@peholderrieth
Mar 3, 2025
Replying to @peholderrieth and @EErives40101
We also had amazing guest lectures from @json_yim (MIT), Ben Burchfiel (Toyota Research Institute), and @cdomingoenrich (Microsoft Research). Thank you so much for supporting us!
4.7K
Peter Holderrieth
@peholderrieth
Mar 3, 2025
Replying to @peholderrieth
@EErives40101 and I did this class together! It was such a fun project to do together! Huge shoutout Ashay Athalye from MIT SOUL for production of the videos and for Tommi Jaakkola for advising us! We also thank everybody at MIT who made this possible! (4/4)
5.4K
Peter Holderrieth
@peholderrieth
Dec 10, 2024
Check out our new Flow Matching guide and codebase! It also includes an extended explanation of Generator Matching with more examples!
Yaron Lipman
@lipmanya
Dec 10, 2024
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/facebookresear…
GIF
arxiv.org
Flow Matching Guide and Code
Flow Matching (FM) is a recent framework for generative modeling that has achieved state-of-the-art performance across various domains, including image, video, audio, speech, and biological...
2.4K
Peter Holderrieth
@peholderrieth
Oct 30, 2024
Replying to @peholderrieth
With GM, we can: 1. Unify and universally characterize Markov models 2. Train them at scale 3. Build multimodal models or combine models via Markov superpositions 4. Train new models such as Euclidean jump models or train diffusion coefficients of a diffusion model (2/5)
2.8K
Peter Holderrieth
@peholderrieth
Oct 30, 2024
Replying to @peholderrieth
Our experiments on image and protein generation show 1. Jump models offer a big unexplored design space with promising results 2. Markov superpositions of jumps + flows improve results We introduce FrameJump, a protein generation method that makes frames jump to a protein! (3/5)
GIF
2K
Peter Holderrieth
@peholderrieth
Nov 17, 2024
Amazing work by an amazing group of researchers! Congrats to the Boltz-1 team!
Gabriele Corso
@GabriCorso
Nov 17, 2024
Thrilled to announce Boltz-1, the first open-source and commercially available model to achieve AlphaFold3-level accuracy on biomolecular structure prediction! An exciting collaboration with @jeremyWohlwend, @pas_saro and an amazing team at MIT and Genesis Therapeutics. A thread!
GIF
2.2K
Peter Holderrieth
@peholderrieth
Oct 30, 2024
Replying to @peholderrieth
This work comes out of an amazing internship hosted by @lipmanya and @rickychen - it truly was a huge pleasure to work together. Big thank you to our amazing co-authors @HavasiMarton, @json_yim , @shaulneta, @itai_gat, Brian Karrer, and (my advisor) Tommi Jaakkola. (5/5)
2.1K