A Flow Matching Algorithm for Many-Shot Adaptation to Unseen Distributions
Tyler Ingebrand, Ruihan Zhao, Kushagra Gupta, David Fridovich-Keil, Sandeep P. Chinchali, Ufuk Topcu · 2026
Read A Flow Matching Algorithm for Many-Shot Adaptation to Unseen Distributions by Tyler Ingebrand, Ruihan Zhao, Kushagra Gupta, David Fridovich-Keil, Sandeep P. Chinchali, Ufuk Topcu (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-2605-06272.txt.
Description
Tyler Ingebrand and five co-authors adapt flow matching generative models to new distributions using many example samples at once. Conditioning output on a target distribution from examples is far less explored than text conditioned output, and retraining for each new target is impractical. Their algorithm, Function Projection for Flow Matching, learns the shared structure of many training distributions ahead of time.
During training, the method learns basis functions spanning the velocity fields of the training distributions. At inference, a new target is handled by a least squares projection onto that basis, with no further training needed. Variants enrich the coefficient calculation, including time dependent coefficients, trading compute for expressivity. Across synthetic and image data sets the approach improved precision and recall over baselines, with the strongest gains on unseen distributions.
Catalog record
| CREATOR | Tyler Ingebrand, Ruihan Zhao, Kushagra Gupta, David Fridovich-Keil, Sandeep P. Chinchali, Ufuk Topcu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.06272