Saved in:
Bibliographic Details
Main Authors: Shlenskii, Daniil, Gushchin, Nikita, Novitskiy, Lev, Dylov, Dmitry V., Korotin, Alexander
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.29920
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911728383557632
author Shlenskii, Daniil
Gushchin, Nikita
Novitskiy, Lev
Dylov, Dmitry V.
Korotin, Alexander
author_facet Shlenskii, Daniil
Gushchin, Nikita
Novitskiy, Lev
Dylov, Dmitry V.
Korotin, Alexander
contents We introduce Midpoint Generative Models (MGM), a principled framework for training one-step generative models. MGM is based on a simple symmetry of Flow Matching with linear interpolation: when the two endpoint distributions coincide, the corresponding drift field vanishes at the midpoint time, $t=1/2$. We show that the norm of this field defines a valid discrepancy between distributions, which we call the Midpoint Divergence. We extend this discrepancy beyond the midpoint by introducing randomly flipped interpolations and further generalize it by replacing deterministic linear Flow Matching interpolations with symmetric stochastic interpolants, yielding a generalized Midpoint Divergence. Finally, we derive a variational formulation of our generalized divergence, yielding a tractable objective for training a one-step generator. The resulting MGM algorithm offers an effective and theoretically grounded approach to generative modeling, achieving competitive performance against existing one-step generative modeling methods.
format Preprint
id arxiv_https___arxiv_org_abs_2605_29920
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Midpoint Generative Models
Shlenskii, Daniil
Gushchin, Nikita
Novitskiy, Lev
Dylov, Dmitry V.
Korotin, Alexander
Machine Learning
We introduce Midpoint Generative Models (MGM), a principled framework for training one-step generative models. MGM is based on a simple symmetry of Flow Matching with linear interpolation: when the two endpoint distributions coincide, the corresponding drift field vanishes at the midpoint time, $t=1/2$. We show that the norm of this field defines a valid discrepancy between distributions, which we call the Midpoint Divergence. We extend this discrepancy beyond the midpoint by introducing randomly flipped interpolations and further generalize it by replacing deterministic linear Flow Matching interpolations with symmetric stochastic interpolants, yielding a generalized Midpoint Divergence. Finally, we derive a variational formulation of our generalized divergence, yielding a tractable objective for training a one-step generator. The resulting MGM algorithm offers an effective and theoretically grounded approach to generative modeling, achieving competitive performance against existing one-step generative modeling methods.
title Midpoint Generative Models
topic Machine Learning
url https://arxiv.org/abs/2605.29920