Posterior Augmented Flow Matching
George Stoica Sayak Paul Matthew Wallingford, Vivek Ramanujan Abhay Nori Winson Han, Ali Farhadi Ranjay Krishna Judy Hoffman · 2026
Read Posterior Augmented Flow Matching by George Stoica Sayak Paul Matthew Wallingford, Vivek Ramanujan Abhay Nori Winson Han, Ali Farhadi Ranjay Krishna Judy Hoffman (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-2605-00825.txt.
Description
George Stoica and eight co-authors improve flow matching, a method that trains a time-dependent field to carry samples from a simple prior to complex data. For high-dimensional images, each training sample guides only one path, so the signal is sparse and noisy. That weakness can cause flow collapse, where varied inputs map to overly similar outputs.
Their Posterior-Augmented Flow Matching replaces single-target guidance with an average over many plausible target completions for each intermediate state. An importance sampling scheme builds that mixture, and the authors prove the new estimator stays unbiased while cutting gradient variance. Across model scales and architectures, image quality improves notably with negligible extra compute.
Catalog record
| CREATOR | George Stoica Sayak Paul Matthew Wallingford, Vivek Ramanujan Abhay Nori Winson Han, Ali Farhadi Ranjay Krishna Judy Hoffman |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 45 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.00825