1905.13229] Private Hypothesis Selection Skip to main content We gratefully acknowledge support from the Simons

2302.07956] Tight Auditing of Differentially Private Machine Learning Skip to main content We gratefully acknowledge support

2406.07407] Private Geometric Median Skip to main content We gratefully acknowledge support from the Simons

2406.07407] Private Geometric Median Skip to main content We gratefully acknowledge support from the Simons

1704.03024] Tight Lower Bounds for Differentially Private Selection Skip to main content We gratefully acknowledge support

PDF Abstract: Differentially private (stochastic) gradient descent is the workhorse of DP private machine learning in

PDF Abstract: Differentially private (stochastic) gradient descent is the workhorse of DP private machine learning in

2006.10559] Differentially-private Federated Neural Architecture Search Skip to main content We gratefully acknowledge support from the

2303.01256] Choosing Public Datasets for Private Machine Learning via Gradient Subspace Distance Skip to main content

2405.20405] Private Mean Estimation with Person-Level Differential Privacy Skip to main content We gratefully acknowledge

2405.20405] Private Mean Estimation with Person-Level Differential Privacy Skip to main content We gratefully acknowledge

2505.23682] Differentially Private Space-Efficient Algorithms for Counting Distinct Elements in the Turnstile Model

2106.13329] Covariance-Aware Private Mean Estimation Without Private Covariance Estimation Skip to main content We gratefully acknowledge support

2311.10825] Pudding: Private User Discovery in Anonymity Networks Skip to main content We gratefully acknowledge

2109.10074] STAR: Secret Sharing for Private Threshold Aggregation Reporting Skip to main content We gratefully acknowledge

2409.09676] Nebula: Efficient, Private and Accurate Histogram Estimation Skip to main content We gratefully acknowledge support

2409.09676] Nebula: Efficient, Private and Accurate Histogram Estimation Skip to main content We gratefully acknowledge support

2109.06153] Relaxed Marginal Consistency for Differentially Private Query Answering Skip to main content We gratefully acknowledge

2106.07153] Iterative Methods for Private Synthetic Data: Unifying Framework and New Methods Skip to main

2110.05679] Large Language Models Can Be Strong Differentially Private Learners Happy Open Access Week from arXiv! YOU make open access

1409.2177] The Large Margin Mechanism for Differentially Private Maximization Skip to main content We

2103.06641] Differentially Private Query Release Through Adaptive Projection Skip to main content We gratefully acknowledge support from

2010.12603] Permute-and-Flip: A new mechanism for differentially private selection Skip to

Nicolas Papernot and 1 other authors View PDF Abstract: For many differentially private algorithms, such as the

advances in auditing which have been used for estimating lower bounds on differentially private algorithms, here we show

the data of a set of people will still be differentially private as long as each

titled Privately Learning Subspaces, by Vikrant Singhal and 1 other authors View PDF Abstract: Private data analysis suffers a

gains but careful comparison between different approaches is challenging. Training is computationally expensive, often done on private datasets of different

use the powerful Sum of Squares method (SoS) to design differentially private algorithms. SoS proofs to

other authors View PDF Abstract: We propose a scheme for auditing differentially private machine learning systems with

a new approach for analyzing the generalization guarantees of differentially private algorithms. Comments: To

Abstract: The permute-and-flip mechanism is a recently proposed differentially private selection algorithm that was

2 other authors View PDF Abstract: A key tool for building differentially private systems is adding Gaussian

invisible actions taken by moderators on Reddit, using a unique dataset of private moderator logs for

to make apparent volunteer labor's value. Using a novel dataset of private logs generated by moderators

based relaxations, however, have several acknowledged weaknesses, either in handling composition of private algorithms or in

DP) is a rigorous notion of data privacy, used for private statistics. The canonical

DP) is a rigorous notion of data privacy, used for private statistics. The canonical

m ds_demo_code/experiment_toolbox/@myCleanup/myCleanup.m ds_demo_code/experiment_toolbox/experiment_load.m ds_demo_code/experiment_toolbox/experiment_run.m ds_demo_code/experiment_toolbox/private/experiment_base.m ds_demo_code/experiment_toolbox

easily misled to generate toxic and biased outputs and leak private information in both

trust the analyzer, local privacy comes at a price: a locally private protocol is less accurate

give the first black-box reduction from privacy to robustness which can produce private estimators with optimal tradeoffs

of computing tight privacy guarantees for the composition of subsampled differentially private mechanisms. Recent algorithms can

of sensitive attribute data. For each domain, we describe how and when private companies collect or infer

of computing tight privacy guarantees for the composition of subsampled differentially private mechanisms. Recent algorithms can

of sensitive attribute data. For each domain, we describe how and when private companies collect or infer

approaches remain a challenge. First, datasets used in existing works are often private and/or

potential for significant harm, particularly from pretraining on biased, obscene, copyrighted, and private information. Emerging ethical approaches

notions have appeared in several recent papers that analyzed composition of differentially private mechanisms. We argue that

used for cross-profile tracking (e.g., linking user behavior across normal and private browsing sessions). Finally, we

and 3 other authors View PDF HTML (experimental) Abstract: Large Language Models (LLMs) often memorize sensitive, private, or copyrighted data during

of a capability to prevent the exfiltration of private data over unauthorized data

Abstract: We initiate a systematic study of algorithms that are both differentially private and run

big data analysis. Although recent works have shown the existence of differentially private sublinear algorithms for

Privacy-protected microdata are often the desired output of a differentially private algorithm since microdata is

essential to secure the communications and devices of private citizens, businesses, and

a decentralized foundation for public social media. It was launched in private beta in February

performance by social group for LMs. The second focuses on risks from private data leaks or LMs

essential to secure the communications and devices of private citizens, businesses, and

to study the impact of clicking on search ads on three popular private search engines which have

easily misled to generate toxic and biased outputs and leak private information in both

easily misled to generate toxic and biased outputs and leak private information in both

x \in \R^N$, we seek to find the differentially private mechanism that has the

give a fast algorithm to optimally compose privacy guarantees of differentially private (DP) algorithms to

and data decisions affect privacy. Overall, our results show that diffusion models are much less private than prior generative models

may be crowdsourced and contain sensitive information. The models should not expose private information in these

Data Structures and Algorithms (cs.DS) [5] arXiv:2311.10825 [ pdf , other ] Title: Pudding: Private User Discovery in

do the same. Tools that use public-key encryption could encode the public/private key-pairs as proquints

et al., 2019 ) , performing arbitrary remote code execution (Zheng & Zhang, 2013 ) , and exfiltrating private data (Ullah et al