Flow Matching for Probabilistic Learning of Dynamical Systems from Missing or Noisy Data

Siddharth Rout, Eldad Haber, Stéphane Gaudreault · 2025

Read Flow Matching for Probabilistic Learning of Dynamical Systems from Missing or Noisy Data by Siddharth Rout, Eldad Haber, Stéphane Gaudreault (2025) 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-2508-01101.txt.

Description

Siddharth Rout, Eldad Haber, and Stéphane Gaudreault wrote this machine learning paper, published in 2025. It forecasts systems that change over time, such as weather or predator and prey counts, when the starting data are noisy or partly missing.

A fuzzy starting point can lead to many plausible futures, so forecasters run crowds of scenarios. Shaking the start with plain random noise often creates impossible states in big systems. The authors instead learn realistic shakes with a flow matching model, a fast generative method. A second flow model then runs each scenario forward with plain equation solving, needing fewer steps than rival diffusion methods. On test beds from animal cycles to world weather data, the approach scored best on a standard forecast skill measure while running much faster.

Catalog record

CREATORSiddharth Rout, Eldad Haber, Stéphane Gaudreault
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE68 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-01101

Browse the full library · About the Tower of Babel