8 papers
Reliable Estimation of Differential Privacy under Selection Bias
Valentin Dorseuil, Meenatchi Sundaram Muthu Selva Annamalai, Jamal Atif +2
Statistical privacy audits are essential for verifying the theoretical privacy guarantees of differentially private (DP) mechanisms. Auditors typically use Membership Inference Att…
Counterfactual Certification for Sequential Unlearning with Long-Run Performance Guarantees
Hanna Benarroch, Jamal Atif, Olivier Cappé
Certified machine unlearning ensures the proximity between the unlearned model and a model retrained without the data to be forgotten. Extending this framework to sequential deleti…
Assessing Per-Sample Membership Inference Vulnerability without Retraining
Valentin Dorseuil, Jamal Atif, Olivier Cappé
Recent work in the privacy literature shows that sample-targeted membership inference attacks (MIAs) significantly outperform untargeted approaches by a wide margin. Motivated by t…
Certified Per-Instance Unlearning Using Individual Sensitivity Bounds
Hanna Benarroch, Jamal Atif, Olivier Cappé
Certified machine unlearning can be achieved through noise injection, with differential privacy guarantees obtained by calibrating the noise to worst-case sensitivity. Yet, the inf…
On the MIA Vulnerability Gap Between Private GANs and Diffusion Models
Ilana Sebag, Jean-Yves Franceschi, Alain Rakotomamonjy +2
Generative Adversarial Networks (GANs) and diffusion models have emerged as leading approaches for high-quality image synthesis. While both can be trained under differential privac…
Memorization in Fine-Tuned Large Language Models
Danil Savine
This study investigates the mechanisms and factors influencing memorization in fine-tuned large language models (LLMs), with a focus on the medical domain due to its privacy-sensit…
Optimal Classification under Performative Distribution Shift
Edwige Cyffers, Muni Sreenivas Pydi, Jamal Atif +1
Performative learning addresses the increasingly pervasive situations in which algorithmic decisions may induce changes in the data distribution as a consequence of their public de…
Differentially Private Gradient Flow based on the Sliced Wasserstein Distance
Ilana Sebag, Muni Sreenivas Pydi, Jean-Yves Franceschi +4
Safeguarding privacy in sensitive training data is paramount, particularly in the context of generative modeling. This can be achieved through either differentially private stochas…