Generalized Leverage Score for Scalable Assessment of Privacy Vulnerability
This provides a computationally efficient method for privacy risk assessment in machine learning, addressing a critical need for scalable vulnerability evaluation without extensive simulations.
The paper tackles the problem of assessing individual data point vulnerability to membership inference attacks without retraining models, showing that risk is governed by influence on the learned model and proposing a generalized leverage score for deep learning, with empirical validation showing strong correlation to attack success.
Can the privacy vulnerability of individual data points be assessed without retraining models or explicitly simulating attacks? We answer affirmatively by showing that exposure to membership inference attack (MIA) is fundamentally governed by a data point's influence on the learned model. We formalize this in the linear setting by establishing a theoretical correspondence between individual MIA risk and the leverage score, identifying it as a principled metric for vulnerability. This characterization explains how data-dependent sensitivity translates into exposure, without the computational burden of training shadow models. Building on this, we propose a computationally efficient generalization of the leverage score for deep learning. Empirical evaluations confirm a strong correlation between the proposed score and MIA success, validating this metric as a practical surrogate for individual privacy risk assessment.