Mean-Shift Distillation for Diffusion Mode Seeking
Vikas Thamizharasan, Nikitas Chatzis, Iliyan Georgiev, Matthew Fisher, Evangelos Kalogerakis, Difan Liu, Nanxuan Zhao, Michal Lukáč · 2025
Read Mean-Shift Distillation for Diffusion Mode Seeking by Vikas Thamizharasan, Nikitas Chatzis, Iliyan Georgiev, Matthew Fisher, Evangelos Kalogerakis, Difan Liu, Nanxuan Zhao, Michal Lukáč (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-2502-15989.txt.
Description
Image generators that build pictures step by step can be shrunk into faster models, but the copy often blurs the most likely outcomes. Vikas Thamizharasan and co-authors offer a 2025 method that keeps those peak outcomes in focus.
Their approach borrows a classic trick for finding peaks in a spread of data. They show the peaks of the trick line up with the true peaks of the model's output spread. A fast sampling scheme makes the method cheap to run. It drops into place of a common older method with no retraining. Tests on made-up and real tasks show better peak match and steadier training, for both picture and three-dimensional output.
Catalog record
| CREATOR | Vikas Thamizharasan, Nikitas Chatzis, Iliyan Georgiev, Matthew Fisher, Evangelos Kalogerakis, Difan Liu, Nanxuan Zhao, Michal Lukáč |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 42 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-15989