MammothDP hardens private boosted trees for trusted hardware
It adds privacy-free early stopping and rollback protection intended to resist timing side channels on TEEs.
Stories per day · 30 days
10 RSS
Relatedprivacy8dp-sgd5federated-learning5privacy-accounting5synthetic-data5diffusion-models2embeddings2graph-privacy2llm2membership-inference2rag2subgraph-counting2alignment1auditing1biomedclip1black-box-evaluation1breast-imaging1celeba1
83 stories, newest first · one card per story
Stories tagged #differential-privacy
Affine subspace constraints improve federated gradient inversion
The authors report better reconstruction than chosen baselines, including when batch size is moderately overestimated.
MOMENT-DP ranks private noise mechanisms using optimizer momentum
Holding curvature, schedule and privacy calibration fixed, changing only momentum can reverse mechanism rankings.
FTP outperforms PTF in differential privacy for functional data
Fit-Then-Protect yields smaller mean squared error than Protect-Then-Fit under Gaussian differential privacy in functional mean estimation
arXiv proposes differentially private framework for GraphRAG
RelGuard is the first differentially private framework for GraphRAG providing formal privacy guarantees.
Audience-conditioned privatization for private text generation
Ext-Classify-Priv framework classifies attributes by audience to preserve only contextually sensitive information in differentially private synthetic data releases.
arXiv releases CoRTeC for differentially private synthetic tabular data
CoRTeC releases synthetic data under differential privacy using noisy counts, decodable without language models or real records.
Interactive LEDP improves subgraph counting accuracy by factor of sqrt n
A two-round interactive algorithm for subgraph counting under local edge differential privacy improves additive error by a factor of sqrt n compared to noninteractive approaches.
Privately assessing data sharing for machine learning
Interactive communication protocol allows two parties to privately test whether their datasets are compatible for joint training under epsilon-differential privacy using Theta(log log m over epsilon) samples for…
Information hierarchy studied for differential privacy filters and odometers
Moving up the information hierarchy yields unbounded utility advantages for privacy filters and arbitrary precision improvements for odometers.
Optimal conversion from zero-concentrated to Gaussian differential privacy
Every ρ-zCDP mechanism satisfies μ_zCDP(ρ)-GDP with sharp bounds √(πρ) ≤ μ_zCDP(ρ) ≤ √(8ρ).
Counting on Dummies enables privacy amplification with truncated batches
Mechanism uses dummy points to privatize batch sizes in DP-SGD while preserving privacy under truncation and subsampling.
Multiscale FFT algorithm improves privacy loss accounting for random allocation
Multiscale FFT reduces runtime from quadratic to O(α^{-1} log(1/α)) and rounding errors to O(α^2/log(t)) for DP-SGD with PREAMBLE.
SPLICE: Differentially Private Synthetic Data with Dimension-Independent Guarantees
SPLICE returns a private distribution with Sliced Wasserstein-1 distance tilde O(1/sqrt(n)) from the target, independent of data dimension d.
Propose-filter-release turns friendly privacy into standard differential privacy
PFR privately filters outliers so noise can scale to typical sensitivity, covering statistics, subspaces, graphs and treatment rules.
Paper derives optimal privacy-budget schedules for private SGD
PB-AMD uses adjoint gradients and entropic updates so noise schedules stay positive and spend the certified zCDP budget.
Noisy aggregation cannot match central DP for private selection
A proposed Sparse Vector Technique primitive reaches central-model error with O(d log-star d) expected queries.
Paper restates robust membership inference for mean estimation
The attacks remain effective when released statistics are highly noisy, for Gaussians and binary products.
Federated breast-imaging models may be confounded by acquisition year
Normal-versus-sick ROC-AUC was 0.940-0.990, but year alone scored 0.974 on the same partition.
Paper privately selects clipping bounds for heavy-tailed data
The method picks a clipping bound by balancing residual coherence against the extra noise variance.
PRISMA builds differentially private synthetic in-car energy traces
At epsilon 0.5, PRISMA beats tabular DP baselines by keeping diverse power profiles instead of collapsing to few modes.
CA-DPSGD uses curvature-aware clipping to reduce DPSGD bias
The method pairs a rectified curvature estimator with anisotropic noise and reports better utility than prior clipping methods.
Paper gives near-optimal private learning of discrete distributions
Polynomial-time pure and approximate DP algorithms match known sample rates up to logarithmic factors for batched user data.
Researchers audit differentially private graph releases
New toolkit falsifies claimed privacy guarantees in Node-DP mechanism.