BLENDER: Enabling Local Search with a Hybrid Differential Privacy Model

By Brendan Avent, Aleksandra Korolova, David Zeber, Torgeir Hovden, Benjamin Livshits. 2017

102 citations recorded by the source corpus.

We propose a hybrid model of differential privacy that considers a combination of regular and opt-in users who desire the differential privacy guarantees of the local privacy model and the trusted curator model, respectively. We demonstrate that within this model, it is possible to design a new type of blended algorithm that improves the utility of obtained data, while providing users with their desired privacy guarantees. We apply this algorithm to the task of privately computing the head of the search log and show that the blended approach provides significant improvements in the utility of the data compared to related work. Specifically, on two large search click data sets, comprising 1.75 and 16 GB, respectively, our approach attains NDCG values exceeding 95% across a range of privacy budget values.

Subfields: Artificial Intelligence, Sociology and Political Science, Privacy & Security, Mathematical Optimization

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Sociology and Political Science

In Privacy & Security

In Mathematical Optimization

Citations in the corpus

Advances and Open Problems in Federated Learning (2019, background): “Hybrid differential privacy Another promising approach is hybrid differential privacy [39], which combines multiple trust models by partitioning users by their trust model preference (e.”
The Power of the Hybrid Model for Mean Estimation (2018, methodology): “We first compare our paper to the closest related work in the hybrid model [3], then discuss other works on DP mean estimation in non-hybrid models, and conclude by discussing other work in hybrid trust models.”
Privacy Amplification by Iteration (2018, background): “: The setting in which some public data from the same distribution as private data is available has been recently identified as promising and practically important [10], [11].”
Differentially-Private "Draw and Discard" Machine Learning (2018, background): “We hypothesize that accuracy and privacy of DDML can also be improved using hybrid approaches to privacy that combine locally private data with public data or data obtained with privacy in the non-local model, as proposed by [19] and [3].”
Collecting and Analyzing Multidimensional Data with Local Differential Privacy (2019, background): “Finally, a recent work [3] introduces a hybrid model that involves both centralized and local differential privacy.”
Connecting Robust Shuffle Privacy and Pan-Privacy (2020, background): “The basic diﬀerential privacy framework is the foundation for many models, including central [18], local [18, 11, 28], pan-[19], blended [6], and most recently shuﬄe [16] privacy.”
AttriGuard: A Practical Defense Against Attribute Inference Attacks via Adversarial Machine Learning (2018, background): “Local differential privacy (LDP): LDP [33–40] is a technique based on ε-differential privacy [41] to protect privacy of an individual user’s data record, i.”
Bypassing the Ambient Dimension: Private SGD with Gradient Subspace Identification (2020, methodology): “[2] Brendan Avent, Aleksandra Korolova, David Zeber, Torgeir Hovden, and Benjamin Livshits.”
Locally Differentially Private Heavy Hitter Identification (2017, background): “[4] propose to combine the centralized and local version of DP together and finds heavy hitters.”
Calibrate: Frequency Estimation and Heavy Hitter Identification with Local Differential Privacy via Incorporating Prior Knowledge (2018, background): “In a hybrid LDP setting [4], some opt-in users share their true items with the data collector, who could obtain prior knowledge about the distribution of the true item frequencies from such opt-in users.”
Privacy at Scale: Local Differential Privacy in Practice (2018, background): “Recent work has proposed a hybrid model where some users follow LDP and some users submit to a trusted aggregator, and both sets are “blended” together [1].”
PROCHLO: Strong Privacy for Analytics in the Crowd (2017, background): “Over the last 3 years, RAPPOR has processed up to billions of daily, randomized reports in a manner that guarantees local differential privacy, without assumptions about users’ trust; similar techniques have since gained increased attention [6,7,70,74].”
A Survey on Differential Privacy for Unstructured Data Content (2022, methodology): “Its original model has been extended to many variants [19] for specific data scenarios, such as Metric DP [11], local DP [20], shuffled DP [5, 33], and hybrid DP [4].”
PrivTrie: Effective Frequent Term Discovery under Local Differential Privacy (2018, background): “Besides, there exist some works [9], [15], [16], [36] focusing on publishing heavy hitters.”