@PPotaptchik

DPhil student at Oxford https://xitter.cf/t.co/JH0l4u7wHv

 Oxford
Joined July 2019
  • Tweets41
  • Following468
  • Followers391
14Media
✨🚀Reward alignment just hit a major milestone🚀✨ Controlling generative models is usually bottlenecked by expensive trajectory rollouts 🐢🐌 or massive particle ensembles 🐢🐌 Meta Flow Maps (MFMs) eliminate both We extend consistency models and flow maps into the stochastic regime—enabling efficient generative control Is that a text-to-image model you see below? Nope. It’s just ImageNet with inference-time steering 🤯🤯🤯 Want unbiased off-policy fine-tuning ? We got that too 📜arxiv.org/abs/2601.14430 🌐meta-flow-maps.github.io 🧵1/10
4
17
1
141
27,196
Peter Potaptchik retweeted
Brian Lee
@BrianLee794309Aug 5
Introducing Beckmann Transport Models (BTMs): a theoretical framework for a new class of generative models under the manifold hypothesis. BTMs admit two complementary instantiations: 1. A one-step transport map learned directly — without jointly learning a velocity field. 2. A time-independent flow whose trajectories halt upon first hitting the data manifold. Paper: arxiv.org/pdf/2608.01692 Tutorial: colab.research.google.com/dr… Joint work with @CoeurdouxF, @PPotaptchik, @msalbergo, @du_yilun, and Eric Vanden-Eijnden. Thanks also to @sophia_tang and Franklin Wang for helpful experimental and presentation discussions.
3
24
3
112
32,995
Peter Potaptchik retweeted
Iskander Azangulov@IAzangulovJun 22
📄 Masked Language Flow Models (MLFMs) We introduce a method to scale flow-based language models to reasoning and instruction-following tasks. Joint work with @kiaashour @leoeleoleo1 @simonvary @prebeschini 🔗 kiaashour.github.io/blog/mas… 1/6
5
8
4
42
12,861
Peter Potaptchik retweeted
Zhihan Yang
@zhihanyang_May 19
📢Excited to share our new paper: Continuous Diffusion Scales Competitively with Discrete Diffusion for Language We introduce RePlaid 🌊, a continuous diffusion language model (DLM) with 🏅Discrete likelihood bound 🏅Scaling laws competitive with SOTA discrete DLMs How? Dive in👇[🧵1/12] Paper: arxiv.org/abs/2605.18530 Work done with my amazing collaborators: @WeiGuo01 @ShuibaiZ69721 @ssahoo_ @YongxinChen1 @ArashVahdat @MardaniMorteza @jwthickstun
6
47
13
234
81,287
Peter Potaptchik retweeted
Oscar Davis
@osclsdMay 19
We were all wondering whether Categorical Flow Maps (CFMs) could scale... 🤔 I couldn't help trying it out... So we scaled CFMs to 1.7B parameters over 2.1T tokens 🚀🔥 Short summary 🧵⬇️
4
31
3
129
17,135
Peter Potaptchik retweeted
Christos Ziakas
@christos_ziakasMay 19
What if a generative policy could generate actions in a single step — and you could optimally align it with Q-guidance in closed form? 📢New paper: Aligning Flow Map Policies with Optimal Q-Guidance We introduce flow map policies + FMQ: a closed-form method for optimally aligning one-step generative policies with Q-guidance. FMQ achieves SOTA in offline-to-online RL across 12 tasks from OGBench and RoboMimic. w/ Alessandra Russo & @bose_joey 🧵 (1/6)
7
13
1
60
8,056
Peter Potaptchik retweeted
Michael Albergo
@msalbergoApr 26
New paper! Presenting Discrete Flow Maps: paper: arxiv.org/abs/2604.09784 blog: malbergo.me/discrete-flow-ma… A laughable problem for me these days is that @nmboffi and I share a research brain, and we have had, time and again, a conversation that ends with “ha so I guess we’re writing the same paper.” Soon we will return to just doing it together :). Here we are doing it again with discrete flow maps and flow language models! A complete and thorough paper led by @PPotaptchik @json_yim @adhisarav @peholderrieth. We took a bit of time to post it to ensure we understood a few more things about the stability of the loss functions. Like @osclsd , @FEijkelboom, and @nmboffi , we think this could be a very helpful paradigm for thinking about fast inference and even better alignment! Here’s our version of the story, and I hope it makes clear how green field this research direction is — we provide a comprehensive picture of the KL losses you can write from the properties of the flow map, some nice geometric proofs about the mean denoiser and the simplex, and find that at this time, the ESD can actually be the most performant, with some caveats. Excited for everyone to work together and push this class of models to their limit!
11
57
9
391
80,682
Peter Potaptchik retweeted
Stat.ML Papers@StatMLPapersApr 2
Adaptive Diffusion Guidance via Stochastic Optimal Control ift.tt/QW9d1h6

Adaptive Diffusion Guidance via Stochastic Optimal Control

Guidance is a cornerstone of modern diffusion models, playing a pivotal role in conditional generation and enhancing the quality of unconditional samples. However, current approaches to guidance...

arxiv.org
9
32
2,348
Peter Potaptchik retweeted
Michael Albergo
@msalbergoFeb 5
Code is now available here: github.com/adh1s/mfm Great work @adhisarav @PPotaptchik !

GitHub - adh1s/mfm: Official Implementation of "Meta Flow Maps enable scalable reward alignment"

Official Implementation of "Meta Flow Maps enable scalable reward alignment" - adh1s/mfm

github.com
Michael Albergo
@msalbergoJan 23
Imagine you could solve an infinite set of transport problems with one Meta Flow Map model that allows you to sample from arbitrary posterior distributions. Now imagine you can do that to construct a really effective estimator of how to adapt a diffusion to solve an RL problem. Now imagine that doing so allows you to even outperform Best-of-N=1000 at a fraction of the compute. Excited to introduce a new paradigm for flow and diffusion models we call Meta Flow Maps, which make this possible 🙂 👾 Learnable with simple modification of existing flow map losses 👾Off-policy fine-tuning algorithm! 👾Extremely effective reward alignment across a variety of rewards for both inference-time steering and learned fine-tuning! arxiv: arxiv.org/abs/2601.14430 project page: meta-flow-maps.github.io/ code: forthcoming Amazing work by @PPotaptchik and @adhisarav to bring these results to life! Really excited about future directions here! Thanks to @yeewhye @AbbasMammadov11 and Alvaro Prat.
1
7
61
6,277
📜 arXiv: arxiv.org/abs/2601.14430 🌐 Project Page: meta-flow-maps.github.io 💻 Code: Forthcoming Huge shoutout to my co-first author @adhisarav! 🚀✨ And a special thanks to our other incredible co-authors @AbbasMammadov11, Alvaro Prat, @msalbergo and @yeewhye Really excited about future directions here! 🧵 10/10

Meta Flow Maps enable scalable reward alignment

Controlling generative models is computationally expensive. This is because optimal alignment with a reward function--whether via inference-time steering or fine-tuning--requires estimating the...

arxiv.org
1
4
5
1,088
Tagging some people who might be interested! @sedielem @ArnaudDoucet1 @du_yilun @HannesStaerk @mmbronstein @ValentinDeBort1 @pabbeel @svlevine @bose_joey @AlexanderTong7 @RickyTQChen @jaschasd @sirbayes @ArashVahdat
3
655
🎆⭐️You thought 2025 would end quietly? Think again. One last exciting update: Tilt Matching!⭐️🎆 We propose a simple, scalable algorithm to sample unnormalized densities and fine-tune generative models 📄: arxiv.org/abs/2512.21829 With @brianlee_lck @msalbergo (1/5)
1
10
1
75
20,345
more replies
But wait, is there cool theory too? Oh boy yes!🤤 We've got: 🔹Connections to SOC 🔹Esscher transform interpretation 🔹Higher-order cumulants 🔹Optimal control variates 🔹Connections to Weighted Flow Matching 🔹Extensions to Flow Maps 📄: arxiv.org/abs/2512.21829 (5/5)

Tilt Matching for Scalable Sampling and Fine-Tuning

We propose a simple, scalable algorithm for using stochastic interpolants to sample from unnormalized densities and for fine-tuning generative models. The approach, Tilt Matching, arises from a...

arxiv.org
1
1
7
1,551
Tagging some people who might be interested! @bose_joey @thjashin @ArnaudDoucet1 @du_yilun @k_neklyudov @AlexanderTong7 @yeewhye @HannesStaerk @mmbronstein @nmboffi @ValentinDeBort1 @JamesTThorn @RickyTQChen @carles_domingo @wgrathwohl @lipmanya
1
10
776
Load more