A large-scale statistical evaluation shows membership inference attacks on LLMs are mostly near-random on average, but a small set of outlier settings are reliably attackable, and performance improves with model size.
This aligns with the obser- vation in the probability density analysis of RUC- AOC scores across dimensions, which confirms the emergent embedding change that we have discov- ered
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
A Statistical and Multi-Perspective Revisiting of the Membership Inference Attack in Large Language Models
A large-scale statistical evaluation shows membership inference attacks on LLMs are mostly near-random on average, but a small set of outlier settings are reliably attackable, and performance improves with model size.