A DUI framework that uses synthetic non-members and mixture proportion estimation to quantify dataset usage in trained models without requiring shadow models or real held-out data.
Low-cost high- power membership inference attacks,
4 Pith papers cite this work. Polarity classification is still indexing.
4
Pith papers citing it
verdicts
UNVERDICTED 4representative citing papers
idSCD uses semantic correlation descriptors to perform dataset membership inference by comparing learned semantic structures, outperforming baselines in NLI, emotion, and medical text experiments.
Attention layers in tabular foundation models enable effective membership inference attacks via pattern concentration, addressed by an inference-time k-anonymity defense on high-risk queries that cuts leakage by ~50% with minimal utility loss.
Authors introduce MLM and CLM specialization methods that avoid memorizing identifiers in sensitive training data while aiming for a privacy-utility tradeoff on medical datasets.
citing papers explorer
-
Dataset Usage Inference without Shadow Models or Held-out Data
A DUI framework that uses synthetic non-members and mixture proportion estimation to quantify dataset usage in trained models without requiring shadow models or real held-out data.
-
idSCD: Identifying Training Datasets through Semantic Correlation Descriptors
idSCD uses semantic correlation descriptors to perform dataset membership inference by comparing learned semantic structures, outperforming baselines in NLI, emotion, and medical text experiments.
-
Privacy Vulnerabilities of Attention Layers in Tabular Foundation Models and Protection of High-Risk Queries
Attention layers in tabular foundation models enable effective membership inference attacks via pattern concentration, addressed by an inference-time k-anonymity defense on high-risk queries that cuts leakage by ~50% with minimal utility loss.
-
Towards the Anonymization of the Language Modeling
Authors introduce MLM and CLM specialization methods that avoid memorizing identifiers in sensitive training data while aiming for a privacy-utility tradeoff on medical datasets.