Knowledge Transfer Scaling Laws for 3D Medical Imaging
Ho Hin Lee, Dongna Du, Chu Wang, Yuankai Huo, Shi Gu, James C. Gee, Yifan Wu · 2026
Read Knowledge Transfer Scaling Laws for 3D Medical Imaging by Ho Hin Lee, Dongna Du, Chu Wang, Yuankai Huo, Shi Gu, James C. Gee, Yifan Wu (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-2605-06859.txt.
Description
Ho Hin Lee and six co-authors derive scaling laws for knowledge transfer in 3D medical imaging, where one foundation model must pretrain across scan types like CT, MRI, and PET. Mixing these domains is now done by heuristic. Yet the authors observe that domains improve at different rates, and transfer is lopsided. Training on one scan type may greatly help another, with little benefit in reverse.
Both reconstruction loss and cross-domain transfer follow predictable power laws, turning data allocation into an optimization problem. The solution reveals hub domains that help many others and island domains needing direct investment. Transfer-aware mixtures beat proportional sampling by up to fifty-eight percent and hold up on disease and segmentation tasks.
Catalog record
| CREATOR | Ho Hin Lee, Dongna Du, Chu Wang, Yuankai Huo, Shi Gu, James C. Gee, Yifan Wu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 57 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.06859