Self-Auditing Parameter-Efficient Fine-Tuning for Few-Shot 3D Medical Image Segmentation

Son Thai Ly, Hien V. Nguyen · 2026

Read Self-Auditing Parameter-Efficient Fine-Tuning for Few-Shot 3D Medical Image Segmentation by Son Thai Ly, Hien V. Nguyen (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-2603-05822.txt.

Description

Ly and colleagues introduce SEA-PEFT, a self-auditing scheme for parameter-efficient fine-tuning in few-shot 3D medical segmentation. Adapting models to clinical sites is hard: domain shift and scarce annotations strain groups lacking engineers. SEA-PEFT treats setup as an allocation task handled during fine-tuning, toggling adapters off to estimate Dice utility. It picks the active set within a parameter budget. On TotalSegmentator and FLARE'22, it improves Dice by more than 2 points over baselines. Across 1, 5, and 10-shot settings it trains under 1 percent of parameters.

Catalog record

CREATORSon Thai Ly, Hien V. Nguyen
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE24812
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.05822

Browse the full library · About the Tower of Babel