MGD: Moment Guided Diffusion for Maximum Entropy Generation
Etienne Lempereur, Nathanaël Cuvelle–Magar, Florentin Coeurdoux, Stéphane Mallat, Eric Vanden-Eijnden · 2026
Read MGD: Moment Guided Diffusion for Maximum Entropy Generation by Etienne Lempereur, Nathanaël Cuvelle–Magar, Florentin Coeurdoux, Stéphane Mallat, Eric Vanden-Eijnden (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-17211.txt.
Description
Etienne Lempereur led a five-author team on generating samples when only partial information is available. Classic maximum entropy methods give sound gauges of doubt from moment constraints, but their sampling via Markov chain or Langevin methods slows sharply in high dimensions. Flow-based models move noise to data fast, yet offer weak promises and can overfit scarce data. The paper presents Moment Guided Diffusion, which blends both lines of work.
Built on the stochastic interpolant setup, the method samples maximum entropy distributions by solving a differential equation. It steers moments to set values in finite time, dodging slow mixing. In the large-volatility limit the authors prove it tends to the maximum entropy distribution and derive an entropy gauge straight from the motion. Tests on finance, turbulence, and cosmology data yield negentropy values for large multiscale flows.
Catalog record
| CREATOR | Etienne Lempereur, Nathanaël Cuvelle–Magar, Florentin Coeurdoux, Stéphane Mallat, Eric Vanden-Eijnden |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 126 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-17211