Aaron Havens

Excited to be at ICML to present our Spotlight paper: Flow Sampling! We propose a simple fixed-point objective for learning diffusion samplers, built on the flow matching marginal construction. More details to come soon! Paper: arxiv.org/abs/2605.03984

If you're in ICML🇰🇷, come checkout #FAS—a natural (yet w/ nontrivial math 🧐) extension of #Adjoint #Sampling to #functional space for sampling irregular time series such as transition paths. 🗓️ Jul 8 (Wed), 5:00–6:45 PM KST 📍Hall A #3408 Great work by Byoungwoo @bw__park 🫡!

Flow Matching obtains its training supervising velocity by conditioning on data samples. But what if you don’t have a dataset, only an unnormalized density? Flow Sampling gets the supervising drift by conditioning on the source instead of the target. Catch @aaronjhavens at #ICML

Couldn't have done it without my coauthors @shaulneta and Brian Karrer. Feel free to reach out or stop by the poster: Thu, Jul 9th, 10:30 AM – 12:15 PM KST HALL A, poster #3502

Excited to be at ICML to present our Spotlight paper: Flow Sampling! We propose a simple fixed-point objective for learning diffusion samplers, built on the flow matching marginal construction. More details to come soon! Paper: arxiv.org/abs/2605.03984

🇰🇷 If you’re at ICML: come check out our Meta FairChem + TUM poster exploring an idea: Speculative decoding, but for simulation trajectories! 🗓️ Tue, Jul 7, 2026 • 2:00 PM – 3:45 PM KST 📍 HALL A #1312 w/ @guennemann, Ray (Meng) Gao, @bwood_m. 🙏 to the entire FAIRChem team!

Stepping away from continuous flows and diffusion for a moment to explore discrete space. Our new #ICML Spotlight introduces ArBGs: framing Boltzmann Generation as an autoregressive, next-token prediction task. Proud of this collaboration! Read the details below. 👇

If you are interested in methods or applications of generative modeling or sampling, sign up for our summer workshop in Cambridge, MA!

⏩ Speculative Sampling For Faster Molecular Dynamics (ICML 2026) How to unlock lossless MD speedups near 10x with a new form of parallelism? Check out the comments 🧵 [1/4] Big shoutout to my Meta internship advisors @bwood_m, Ray Gao and my academic advisor @guennemann!

@liyzhen2 If you want to fine tune / tilt a pretrained flow-matching model by a reward, it can be formulated as learning the appropriate tilted source distribution: arxiv.org/abs/2508.14807…

Tons of papers re diffusion/flow matching at ML confs these days, but to my surprise very few of them consider learning the prior🤔 Am I missing any important work here? 🙏 for suggestions

🚀MIT Flow Matching and Diffusion Lecture 2026 Released (diffusion.csail.mit.edu)! We just released our new MIT 2026 course on flow matching and diffusion models! We teach the full stack of modern AI image, video, protein generators - theory and practice. We include: 📺 Videos: Step-by-step derivations. 📝 Notes: Mathematically self-contained lecture notes 💻 Coding: Hands-on exercises for every component We fully improved last years’ iteration and added new topics: latent spaces, diffusion transformers, building language models with discrete diffusion models. Everything is available here: diffusion.csail.mit.edu A huge thanks to Tommi Jaakkola for his support in making this class possible and Ashay Athalye (MIT SOUL) for the incredible production! Was fun to do this with @RShprints! #MachineLearning #GenerativeAI #MIT #DiffusionModels #AI

Presenting our spotlight paper on trust regions for optimal control at NeurIPS, arxiv.org/pdf/2508.12511. We show that KL-equipspaced measure transport can be interpreted as geometric annealing with adaptive step sizes, leading to major performance gains on hard control problems.

I'll present a highly-scalable reward-driven diffusion sampler, #Adjoint #Schrödinger #Bridge #Sampler, today at #NeurIPS. Come by our oral talk or poster session to chat in person! 🗣️ Oral talk: 10-10:20am, Upper Level Ballroom 20D 🧾 Poster: 11am-2pm, #612 🔗 Slides: shorturl.at/ZfzeP 💻 Code: shorturl.at/96xfY

New work: “GLASS Flows: Transition Sampling for Alignment of Flow and Diffusion Models”. GLASS generates images by sampling stochastic Markov transitions with ODEs - allowing us to boost text-image alignment for large-scale models at inference time! arxiv.org/pdf/2509.25170 [1/7]

Sampling is hard b/c target distribution can be high-dim with many modes. ML can help, even when state space is discrete (thus non-differentiable)! arxiv.org/abs/2508.10684 constructs a strong sampler by fine tuning a discrete diffusion model via stochastic optimal control / RL.

Georgia Tech AI4Science Center is soft launched, and I'm excited to be an Associate Director. ai4science.ai.gatech.edu Collaboration+Participation of all kinds are welcomed. Please get in touch! Thanks to @gtsciences for supports. Retweets appreciated! @GeorgiaTech #AI4Science

Adjoint-based diffusion samplers have simple & scalable objectives w/o impt weight complication. Like many, though, they solve degenerate Schrödinger bridges, despite all being SB-inspired. 📢 Proudly introduce #Adjoint #Schrödinger #Bridge #Sampler, a full SB-based sampler that is simple to implement, scalable, practically very effective, theoretically sounded, and extends AM beyond memoryless noise schedule. Great collab w/ @jaemoo51133 @RickyTQChen @bkmi13 @YongxinChen1 🙌🙌 arxiv.org/abs/2506.22565

Excited to share Quetzal, a simple but scalable model for building 3D molecules atom-by-atom. 🐉 Named after Quetzalcoatl, the Aztec god of creation We equip a standard causal transformer with a per-atom diffusion MLP to model the continuous 3D position of the next atom. [1/3]

Reward-driven algorithms for training dynamical generative models significantly lag behind their data-driven counterparts in terms of scalability. We aim to rectify this. Adjoint Matching poster @cdomingoenrich Sat 3pm & Adjoint Sampling oral @aaronjhavens Mon 10am FPI

We are presenting 3 orals and 1 spotlight at #ICLR2025 on two primary topics: On generalizing the data-driven flow matching algorithm to jump processes, arbitrary discrete corruption processes, and beyond. And on highly scalable algorithms for reward-driven learning settings.

This work was done during a PhD internship at FAIR NYC, thanks to my amazing supervisors @brandondamos , @RickyTQChen and Brian Karrer. Special thanks to our core contributors: @bkmi13 @bingyan4science @xiangfu_ml @guanhorng_liu (and of course @cdomingoenrich)

Our evaluation offers a new challenging sampling benchmark for molecular conformer generation. The benchmark features real, drug-like molecules from the SPICE dataset, and we hope it drives direct and tangible progress in sampling for computational chemistry (coming soon).

This lets us train conditional diffusion samplers directly from expensive energy functions, namely, state-of-art molecular foundation models, amortizing sampling across thousands of molecules—unlike traditional samplers, which require heavy energy access per new molecule sample.