AMNESIA: A Large Scale Medical Unlearning Benchmark Suite with Disease-Informed Analysis

Saeedeh Davoudi, Reihaneh Iranmanesh, Ophir Frieder, Nazli Goharian · 2026

Read AMNESIA: A Large Scale Medical Unlearning Benchmark Suite with Disease-Informed Analysis by Saeedeh Davoudi, Reihaneh Iranmanesh, Ophir Frieder, Nazli Goharian (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-2605-30599.txt.

Description

Davoudi and colleagues present AMNESIA, the first large-scale open benchmark for making medical language models selectively forget. Medical knowledge keeps changing, so trained models must drop outdated facts without full retraining, yet existing unlearning tests use synthetic or small general data. AMNESIA holds over seventy thousand question-answer pairs from nearly nine thousand patient notes across eleven disease groups, mixing factual recall and clinical reasoning questions. Tests of four unlearning methods show that forgetting one patient erodes knowledge of others with the same condition.

Catalog record

CREATORSaeedeh Davoudi, Reihaneh Iranmanesh, Ophir Frieder, Nazli Goharian
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE55091
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.30599

Browse the full library · About the Tower of Babel