The paper presents an information‑theoretic framework that derives a distortion bound for reconstruction attacks on models trained with DP‑SGD. The bound can be computed before training and applies even when an adversary observes the training trajectory and knows all other training data. It separates the information leaked by training (conditional mutual information) from the inherent uncertainty in the data (source entropy) and uses rate‑distortion theory to produce a lower bound on expected reconstruction error, instantiated in both pixel and LPIPS feature spaces.
arXiv:2609.37344v1 Announce Type: cross Abstract: Data reconstruction attacks have empirically been successful in recovering training samples from learned models, raising privacy concerns and motivat...
arXiv:2607. 12354v1 Announce Type: new Abstract: In this paper, we challenge the prevailing view that information dependency (including rote memorization) drives training data exposure to image reconstruction attacks.
In this paper, we show that standard evaluations of high-resolution Model Inversion Attacks (MIAs) significantly underestimate training-data privacy leakage. State-of-the-art privacy defenses, standar...
arXiv:2610.07677v1 Announce Type: new Abstract: In this paper, we show that standard evaluations of high-resolution Model Inversion Attacks (MIAs) significantly underestimate training-data privacy le...
arXiv:2601. 14033v2 Announce Type: replace Abstract: Machine learning models are increasingly served behind APIs.
arXiv:2409. 01062v4 Announce Type: replace Abstract: Model Inversion (MI) attacks pose a significant privacy threat by reconstructing private training data from machine learning models.
arXiv:2503. 10945v3 Announce Type: replace-cross Abstract: Current practices for reporting differential privacy (DP) guarantees for machine learning (ML) algorithms such as DP-SGD provide an incomplete and potentially misleading picture.
The paper compares four audit methods for assessing identity‑level differential privacy in pre‑trained, black‑box face generators. Each method—GaussMech, KDE‑LR, MMD‑TV, and ROC‑HT—has distinct assumptions, hyperparameters, and finite‑sample limitations, and they produce markedly different epsilon estimates when applied to FaceFusion and InstantID. The study finds that all methods reveal significant identity distinguishability, but none can be reliably ranked in this high‑distinguishability regime, suggesting that future work should evaluate them on partially private mechanisms.
arXiv:2605. 06259v3 Announce Type: replace Abstract: Within the $f$-DP framework, we derive a tight analysis of the trade-off function for Differentially Private Stochastic Gradient Descent (DP-SGD) with subsampling based on random allocation in which each sample is independently assigned to exactly one of $M$ minibatches per epoch, each minibatch corresponding to one of the $M$ SGD rounds within a single epoch.
arXiv:2608.28934v1 Announce Type: new Abstract: Differential privacy (DP) has traditionally been used to provide theoretical upper bounds on an algorithm's stability to changing its training data. In...
arXiv:2608.23382v2 Announce Type: replace Abstract: Instance encoding is a popular empirical technique for privacy enhancement when sharing data to an untrusted server. It transforms sensitive data t...
arXiv:2601. 10237v3 Announce Type: replace Abstract: Differentially Private Stochastic Gradient Descent (DP-SGD) is the dominant paradigm for private training, but its fundamental limitations under worst-case adversarial privacy definitions remain poorly understood.