Collecting and Analyzing Data from Smart Device Users with Local Differential Privacy

By Thông T. Nguyên, Xiaokui Xiao, Y. Yang, Siu Cheung Hui, Hyejin Shin, Junbum Shin. 2016

215 citations recorded by the source corpus.

Organizations with a large user base, such as Samsung and Google, can potentially benefit from collecting and mining users' data. However, doing so raises privacy concerns, and risks accidental privacy breaches with serious consequences. Local differential privacy (LDP) techniques address this problem by only collecting randomized answers from each user, with guarantees of plausible deniability; meanwhile, the aggregator can still build accurate models and predictors by analyzing large amounts of such randomized data. So far, existing LDP solutions either have severely restricted functionality, or focus mainly on theoretical aspects such as asymptotical bounds rather than practical usability and performance. Motivated by this, we propose Harmony, a practical, accurate and efficient system for collecting and analyzing data from smart device users, while satisfying LDP. Harmony applies to multi-dimensional data containing both numerical and categorical attributes, and supports both basic statistics (e.g., mean and frequency estimates), and complex machine learning tasks (e.g., linear regression, logistic regression and SVM classification). Experiments using real data confirm Harmony's effectiveness.

Subfields: Artificial Intelligence, Computer Science Applications

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

1. Philip S. Yu
2. Ninghui Li
3. Badih Ghazi
4. Ravi Kumar
5. Lichao Sun
6. Tianhao Wang
7. Somesh Jha
8. Rasmus Pagh

In Computer Science Applications

1. Ninghui Li
2. Xiaokui Xiao
3. Divesh Srivastava
4. Tianhao Wang
5. Philip S. Yu
6. Jun Zhao
7. Kui Ren
8. Ayfer Özgür

Citations in the corpus

• Breaking the Communication-Privacy-Accuracy Trilemma (2020, background): “The recent works of [32] and [33] also consider estimating empirical mean under ε-LDP.”
• Marginal Release Under Local Differential Privacy (2017, background): “m 0:2 (higher privacy) to 1:4 (lower privacy). Note that the theory shows that and Nare tightly related: decreasing means Nmust be increased to obtain the same accuracy. Some prior work on LDP e.g. [29] studies a smaller regime of values, at the expense of a much larger user population. We begin ou”
• Federated Model Distillation with Noise-Free Differential Privacy (2020, critical): “Some approaches that also considered LDP into FL can only support shallow models such as logistic regression and only focus on simple tasks and datasets (Nguyên et al. 2016; Wang et al. 2019; Zhao et al. 2020).”
• Locally Differentially Private Data Collection and Analysis (2019, background): “However, the works [39, 46] have indicated that it will degrade the accuracy of the learning algorithms by iterating one tuple multiple times in the local private setting.”
• Privacy at Scale: Local Differential Privacy in Practice (2018, methodology): “This approach has been proposed for building machine learning models [19].”
• CALM: Consistent Adaptive Local Marginal for Marginal Release under Local Differential Privacy (2018, background): “But in the local setting, several previous work pointed out that by partitioning users into groups, and having each group use the full privacy budget, the overall error will be smaller [26, 37, 38].”
• Locally Differentially Private Protocols for Frequency Estimation (2017, background): “Samsung proposed a similar system [21] which enables collection of not only categorical answers (e.”
• Locally Differentially Private Heavy Hitter Identification (2017, background): “Samsung proposed a similar system [26] which enables collection of not only categorical answers (e.g., screen resolution) but also numerical answers (e.g., time of usage, battery volume), although it is not clear whether this has been deployed by Samsung.”
• Privacy and Robustness in Federated Learning: Attacks and Defenses (2020, background): “Other approaches that also considered to apply LDP to FL can only support shallow models such as logistic regression and only focus on simple tasks and datasets [83], [84], [85].”
• LDP-FL: Practical Private Aggregation in Federated Learning with Local Differential Privacy (2020, background): “Some of them [Nguyên et al., 2016; Seif et al., 2020] only focus on small tasks and simple datasets and do not support deep learning approaches yet.”
• Private Heavy Hitters and Range Queries in the Shuffled Model (2019, methodology): “The Hadamard response was also used in [17, 18, 41] for similar applications, namely private frequency estimation.”
• Consistent and Accurate Frequency Oracles under Local Differential Privacy (2019, methodology): “Mean estimation is also a building block in LDP, but most of them transform the numerical value to a discrete value using stochastic round, and then apply frequency oracles [10, 12, 13, 25].”
• Heavy Hitter Estimation over Set-Valued Data with Local Differential Privacy (2016, methodology): “One key idea in the LDPMiner building blocks, inspired by another technique [27] for multi-dimensional data analysis under LDP, is that each user does not report all items; instead, she chooses a random item and reports it.”