🔥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

Flows and transport methods are widely used to connect one distribution to another. What about going up one level to transporting between distributions over distributions? My new talk overviews some methods in this space: bamos.github.io/presentations/… & a quick 🧵 below

🧬What are protein language models (PLMs) actually learning about biology? Our paper introduces InterPLM - a framework that reveals interpretable features in PLMs using sparse autoencoders, giving us a window into how these models represent protein structure and function. 🧵(1/9)

A common question nowadays: Which is better, diffusion or flow matching? 🤔 Our answer: They’re two sides of the same coin. We wrote a blog post to show how diffusion models and Gaussian flow matching are equivalent. That’s great: It means you can use them interchangeably.

Come to Cambridge! There is a tenure-track position at Kempner and Harvard CS. Please share around: academicpositions.harvard.edu/postings/14362

Wow, thanks for the kind words Nando de Freitas ! Check out Willis (Nanye) Ma who really drove this paper and of course the wonderful Mark Goldstein Nicholas Boffi Saining Xie on X. (Mark Goldstein is also on the job market!)

Rare sincere tweet: December can be tough in academia. As a student I thought everyone had it together. As an advisor you see that is very much not true. Generally, at least as a starting place, a really recommend finding someone who you can go on a long walk with to talk it

So tired of seeing Wikipedia bashing. The most egregious instance of this was OpenAI employees claiming their latest release obsoletes the project. Such remarkable hubris. Makes me want to tithe my income to Wikipedia, an extraordinary contribution to humanity. Donate if you can.

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!

We have a new cool preprint with Cheuk Kit Lee. We developed a sweet Sequential Monte Carlo algorithm for unbiased samples from a tempered distribution p(x0)p(y|x0)^α and applied it to discrete diffusion for text arxiv.org/abs/2502.06079. Huge thanks to Francisco Vargas Michael Albergo

Having trouble sampling from a complex distribution? Peter Holderrieth, Michael Albergo, and #JameelClinic PI Tommi Jakkola introduce LEAPS, which samples values from discrete distributions more efficiently & at scale via continuous-time Markov chains. ⭐️Paper: arxiv.org/pdf/2502.10843

& here are some of my favorite papers on the unification of flows and diffusion. What a decade!! (From my presentation here: bamos.github.io/presentations/…)

Excited to be at iclrconf for #ICLR2025! I’ll give a talk at the Frontiers on Probabilistic Inference workshop to discuss work with Eric Vanden-Eijnden, Peter Holderrieth, Cheuk Kit Lee, Paul Jeha, and Francisco Vargas! Let me know about your work, I will come by :) sites.google.com/view/fpiworksh…

Congratulations to Peter Holderrieth Michael Albergo @ICLR2025 and Tommi Jaakkola for winning the best paper award for their work entitled "LEAPS: A discrete neural sampler via locally equivariant networks" at this year's Frontiers in Probabilistic Inference workshop #ICLR2025!

Francisco Vargas and Michael Albergo with a combined talk on sampling inference and transport at #FPIworkshop.

#FPIworkshop best paper award goes to Peter Holderrieth Michael Albergo and Tommi Jaakkola. Congrats and great talk Peter!

What is the probability of an image? What do the highest and lowest probability images look like? Do natural images lie on a low-dimensional manifold? In a new preprint with Zahra Kadkhodaie Eero Simoncelli, we develop a novel energy-based model in order to answer these questions: 🧵