Grouping-Based Reliable Privacy Preservation for Blockchain-Assisted Data Aggregation in Mobile Crowdsensing

Abstract

Privacy-preserving data aggregation is an important technology for mobile crowdsensing. Blockchain-assisted data aggregation enables the traceability of sensing data to improve the trustworthiness of data aggregation results. However, directly using blockchains for data aggregation may introduce the risk of privacy leakage because all nodes, including malicious nodes, can access the data on blockchains. In this paper, we propose a grouping-based reliable privacy-preserving data aggregation (RPPDA) method using private blockchains for mobile crowdsensing. First, the sensing nodes are divided into multiple groups, and each group maintains a private blockchain to store the data aggregation records, which avoids the leakage of the aggregated results and ensures the traceability of the sensory data. Then, a zero-sum noise-adding mechanism is utilized to not only preserve the private information during aggregation and ensure the correctness of the aggregated results but also improve the efficiency of privacy preservation. Furthermore, we theoretically prove the correctness, privacy, efficiency, and reliability of the proposed RPPDA algorithm. Real-world and simulated experiments demonstrate the effectiveness and advantages of the proposed RPPDA algorithm in terms of correctness, efficiency, and privacy.

Optimizing recommender system using federated clustering

“…Additionally, the K-Fed algorithm needs to take raw user information and process it during the client-server exchange, thus exposing sensitive personal data (such as user health) to the server through model inversion attacks. [8][9][10][11]. In order to solve the above-mentioned difficulties, this paper schedules a method to enlarge centroid assignment and keep the leakage of customer center.…”

Towards Efficient Federated Learning Using Agile Aggregation in Internet of Vehicles

“…High-dynamic vehicles in IoV only need to transmit the trained model parameters to road side units (RSUs) by federated learning. It does not need to share the entire original dataset, which reduces the risk of privacy leakage [14,19,20]. Vehiclesensed data are trained locally, and only the trained model parameters are transmitted to the server, which improves privacy preservation during the learning process.…”