Mohsin Hasan

📢Mar 16 (Mon): Discrete Feynman-Kac Correctors 🤔Discrete diffusion models are powerful, but out of the box they give little control over the target distribution!! 🔑Discrete Feynman-Kac Correctors fix this by using Sequential Monte Carlo (SMC) to modify the distribution by - Annealing - Composing multiple models, or - Tilting with external reward functions. All at inference time with no retraining needed! 💡This unlocks things like boosting coding performance, sampling across a range of temperatures in the Ising model, and generating higher quality protein sequences. This Monday, Mohsin Hasan (Université de Montréal, Mila) (hasanmohsin.github.io) and Viktor Ohanesian (Imperial College London) (@OhanesianViktor, scholar.google.com/citations?user…) will co-present their jointly led paper Discrete Feynman-Kac Correctors. Collaborators: Artem Gazizov (Harvard), @Yoshua_Bengio, @A_Aspuru_Guzik, Roberto Bondesan (Imperial College London), @martoskreto, @k_neklyudov Paper link: arxiv.org/abs/2601.10403

8/8 This is joint work with a great team: @OhanesianViktor * (equal contributor), @ag27182, @Yoshua_Bengio, @A_Aspuru_Guzik, Roberto Bondesan, @martoskreto, @k_neklyudov For more, check out: 📄 Paper: arxiv.org/abs/2601.10403 💻 Code: github.com/hasanmohsin/di…

7/8 Finally, we apply DFKC to sample from reward-tilted distributions for proteins. This lets us generate higher quality proteins as measured by external rewards such as ESM2 likelihood and Thermostability, compared to standard guidance methods.

🚀New paper: Discrete Feynman-Kac Correctors! It’s an inference time method that modifies discrete diffusion models via annealing, products, & reward-tilting. W/o training, it generalizes past the Ising critical temp. and boosts coding performance! arxiv.org/abs/2601.10403 🧵👇

Come check out our poster this Wednesday at 4:30pm @icmlconf !! Happy to chat about diffusion, GFlowNets and RL!

Is there a universal strategy to turn any generative model—GANs, VAEs, diffusion models, or flows—into a conditional sampler, or finetuned to optimize a reward function? Yes! Outsourced Diffusion Sampling (ODS) accepted to @icmlconf , does exactly that!

@lostinio @miniapeur @VladZamfir I think most force interactions between particles are 2-body (since that + conservation of momentum is basically equivalent to Newton's 3rd law). Apparently though there can be 3 body forces for the strong force, and I've heard the term "chiral 3-body interaction" before

@miniapeur @VladZamfir Some external forces probably yes. But electrons for example interact with each other exchanging photons. It is one to one interaction. Why nature isnot hypergraph in this case?

Graph, simplicial complex, hypergraph. Graphs are convenient in a range of applications. However, they can only represent dyadic interaction (relations between two vertices) by means of edges. (1/4)