Weighting What Matters: Boosting Sample Efficiency in Medical Report Generation via Token Reweighting
Alexander Weers 1,2 [email protected], Daniel Rueckert 1,2,3 [email protected], Martin J. Menten 1,2 [email protected], 1 TUM School of Computation, Information and Technology, Technical University of Munich, DE 2 Munich Center for Machine Learning (MCML), DE 3 Department of Computing, Imperial College London, UK · 2026
Read Weighting What Matters: Boosting Sample Efficiency in Medical Report Generation via Token Reweighting by Alexander Weers 1,2 [email protected], Daniel Rueckert 1,2,3 [email protected], Martin J. Menten 1,2 [email protected], 1 TUM School of Computation, Information and Technology, Technical University of Munich, DE 2 Munich Center for Machine Learning (MCML), DE 3 Department of Computing, Imperial College London, UK (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-2604-21082.txt.
Description
Alexander Weers, Daniel Rueckert, and Martin Menten tackle a shortage of labeled medical data for systems that draft radiology-style reports from images. Standard training punishes every wrong word equally, so they weight the loss toward words that carry clinical meaning. On eye-care reports, the reweighted training matched standard quality while using up to ten times less training data, a gain that held across several data sizes.
Catalog record
| CREATOR | Alexander Weers 1,2 [email protected], Daniel Rueckert 1,2,3 [email protected], Martin J. Menten 1,2 [email protected], 1 TUM School of Computation, Information and Technology, Technical University of Munich, DE 2 Munich Center for Machine Learning (MCML), DE 3 Department of Computing, Imperial College London, UK |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 9728 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.21082