The Amazing Stability of Flow Matching

Rania Briq [email protected], Michael Kamp, Ohad Fried, Sarel Cohen, Stefan Kesselheim · 2026

Read The Amazing Stability of Flow Matching by Rania Briq [email protected], Michael Kamp, Ohad Fried, Sarel Cohen, Stefan Kesselheim (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-2604-16079.txt.

Description

Briq and four colleagues investigate what drives quality and diversity in flow-matching models: architecture or training datasets. On CelebA-HQ, flow matching stays stable even after pruning 50 percent of the dataset, preserving quality and diversity. Pruning barely shifts the latent representation: full-data and pruned-data models map the same seed to visually similar outputs. Similar stability holds when changing architecture or training setup, with the latent representation maintained throughout. The results quantify this stability and help explain why flow-matching models prove reliable under perturbations.

Catalog record

CREATORRania Briq [email protected], Michael Kamp, Ohad Fried, Sarel Cohen, Stefan Kesselheim
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE17867
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.16079

Browse the full library · About the Tower of Babel