GRASP: Guided Residual Adapters with Sample-wise Partitioning
Felix Nützel, Mischa Dombrowski, Bernhard Kainz, [0.25em] Friedrich-Alexander-Universität Erlangen-Nürnberg · 2025
Read GRASP: Guided Residual Adapters with Sample-wise Partitioning by Felix Nützel, Mischa Dombrowski, Bernhard Kainz, [0.25em] Friedrich-Alexander-Universität Erlangen-Nürnberg (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-2512-01675.txt.
Description
Felix Nutzel, Mischa Dombrowski, and Bernhard Kainz lift image generation for rare diseases. Text-to-image diffusion models make fine images for common prompts. In medical use, rare conditions form a small tail of data. The tail output then collapses in quality and variety.
Their GRASP method adds guided residual adapters with sample-wise splits. Adapters steer the model toward tail cases. Better tail images can augment scarce training sets. That helps diagnostic tools learn rare pathologies they would else miss in clinics with few real cases. Adapters are small add-ons, so clinics can tune the base model for their own rare cases without retraining the whole network.
Catalog record
| CREATOR | Felix Nützel, Mischa Dombrowski, Bernhard Kainz, [0.25em] Friedrich-Alexander-Universität Erlangen-Nürnberg |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 46 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-01675