Fine-tuning medical LLMs with DP-SGD on MIMIC-III top-50 ICD codes cuts micro-F1 by more than 40% and widens the gender recall gap by roughly 3 percentage points.
Membership inference attacks from first principles
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Can large language models be privacy preserving and fair medical coders?
Fine-tuning medical LLMs with DP-SGD on MIMIC-III top-50 ICD codes cuts micro-F1 by more than 40% and widens the gender recall gap by roughly 3 percentage points.