Flow Sampling: Learning to Sample from Unnormalized Densities via Denoising Conditional Processes

Aaron Havens, Brian Karrer, Neta Shaul · 2026

Read Flow Sampling: Learning to Sample from Unnormalized Densities via Denoising Conditional Processes by Aaron Havens, Brian Karrer, Neta Shaul (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-2605-03984.txt.

Description

Aaron Havens, Brian Karrer, and Neta Shaul introduce Flow Sampling, a method for drawing samples from distributions known only through an unnormalized energy function. Unlike generative modeling, no data samples exist; the target is defined by a formula, and evaluating the energy is often costly, so an efficient learned sampler is valuable. Their framework builds on diffusion models and flow matching for the data-free setting.

The training objective conditions on a noise sample and matches a denoising diffusion drift built from the energy function, the mirror image of the usual one. An interpolant process minimizes energy evaluations during training. The method extends naturally to Riemannian manifolds, with a closed-form conditional drift derived for constant-curvature spaces including hyperspheres and hyperbolic spaces, and tests run from synthetic energies to molecular conformer generation.

Catalog record

CREATORAaron Havens, Brian Karrer, Neta Shaul
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE77 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.03984

Browse the full library · About the Tower of Babel