|Title:||Highdimensional data collection with local differential privacy : benchmark and optimizations|
|Advisors:||Hu, Haibo (EIE)|
Hong Kong Polytechnic University -- Dissertations
|Department:||Department of Electronic and Information Engineering|
|Pages:||56 pages : color illustrations|
|Abstract:||Local differential privacy (LDP),which perturbs user's data locally and only sends the noisy version of her information to the aggregator, is a novel and prospective privacy-preserving framework. In LDP, the data collector, which collects and analyzes perturbed data from users, has no access to original data. Thus, privacy is protected from its source. However, a primary drawback of LDP is that most works are strictly limited to lowdimensional scenarios and they could face serious challenges if directly extended to high-dimensional settings, no matter how excellent the algorithm is in low dimensions. In this paper, we first present a comprehensive review on high-dimensional LDP and provide corresponding optimal settings by theoretical analysis. Next, an evaluation benchmark referred as LDP-Bench, which efficiently takes advantage of the peculiarity of highdimensional data, is elaborately designed for impartially parallel comparisons between LDP mechanisms. Then, we conduct stress tests on three representative LDP mechanisms on this benchmark to confirm its effectiveness and generality. To further address the poor utility of Duchi et al.'s solution, an unbiased aggregation method is proposed by introducing an iterative machine learning algorithm, which is shown to achieve impressively higher utility than Duchi et al.'s solution. Further, we are the first to extend Duchi et al.'s solution to implement frequency estimation. Finally, experimental results based on our benchmark confirm the effectiveness of these strategies.|
|Rights:||All rights reserved|
Files in This Item:
|5656.pdf||For All Users (off-campus access for PolyU Staff & Students only)||424.42 kB||Adobe PDF||View/Open|
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item: