HOFAR: High-Order Augmentation of Flow Autoregressive Transformers
Yingyu Liang, Zhizhou Sha, Zhenmei Shi, Zhao Song, Mingda Wan · 2025
Read HOFAR: High-Order Augmentation of Flow Autoregressive Transformers by Yingyu Liang, Zhizhou Sha, Zhenmei Shi, Zhao Song, Mingda Wan (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-2503-08032.txt.
Description
Yingyu Liang and four co-authors present HOFAR in a 2025 paper on image generation. Recent systems blend flow methods with transformer networks to make sharp images. One leading blend, FlowAR, still tracks only the simplest path shape during generation, which limits quality.
HOFAR adds higher-order guidance to that blend. The extra terms describe how the generation path bends and changes, not just where it points. Theory and tests in the paper show clear gains over the base model. The authors frame the method as a general way to study path behavior in this model family. Better path tracking may lift image quality in later systems too.
Catalog record
| CREATOR | Yingyu Liang, Zhizhou Sha, Zhenmei Shi, Zhao Song, Mingda Wan |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 39 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-08032