Choosing an appropriate division method is crucial for partitioning two-dimensional spatial data under the constraints of differential privacy. The current mainstream partitioning methods include grid-based partitioning and hierarchical partitioning. In order to optimize query accuracy while satisfying differential privacy conditions, it remains challenge to achieve the sum minimization of noise error and uniformity assumption error. To address this issue, we propose the HOLG (Hierarchical Optimization of Logical Grids) algorithm, employing a ”divide-merge-divide” approach. It begins with fine-grained grid partitioning of the data domain, followed by heuristic merging of grids with similar data distributions. After determining the scale of the query domain, the merged regions are further subdivided into smaller regions with similar query probabilities, constructing a hierarchical structure to reduce uniformity assumption errors. Additionally, we design a novel noise injection method and introduce consistency constraints to further minimize noise errors. To reduce the time complexity of the HOLG partitioning method, Huffman trees is employed to optimize the processing of the hierarchical tree set generated by HOLG, ensuring query utility while effectively reducing the query response time for the partitioning algorithm. Experimental results on large-scale spatial datasets demonstrate that HOLG outperforms similar algorithms in query accuracy. Furthermore, when combined with the Huffman tree optimization, it effectively reduces query response time.

The widespread use of smartphones has yielded a wealth of behavioral sequence data from user interactions. These interactions offer insights into user preferences and patterns for personalized behavior prediction. However, there are some challenges in current privacy-preserving works for analyzing these data. These approaches have suboptimal service quality with smaller but longer datasets and insufficient emphasis on secure pattern storage and retrieval in real-world applications. To handle these challenges on smartphones, we propose a novel Privacy-enhanced Frequent Sequence Mining and Retrieval (PrivFSMR) framework for this scenario. Specifically, we first introduce a dynamic sequence truncation to anonymize the maximum sequence length of datasets. Following this, we design a privacy-enhanced FSM algorithm to uncover patterns, effectively reducing the privacy budget by integrating differential privacy and the Markov assumption. During the secure pattern storage, PrivFSMR employs symmetric encryption for protection and constructs an encrypted index forest for retrieval. Lastly, future behavior retrieval leverages current device information and the index forest to search similar patterns, thereby predicting potential user behaviors in the future. A comprehensive security analysis proves the PrivFSMR framework guarantees differential privacy and maintains storage and retrieval confidentiality in the lifecycle. In addition to using two publicly available datasets, we also collected a real behavior dataset within 2-4 weeks from 30 users for evaluation. Experimental results on three datasets demonstrate that PrivFSMR excels in mining frequent patterns and predicting future behaviors compared to existing approaches.

With the growing popularity of Internet of Things (IoT) systems, device-free action recognition begins to attract extensive attention due to its friendly feasibility in broad applications, such as human–computer interaction and smart elderly care. Considering abundant information in the vision modality, existing methods adopt the cross-model methods for performance enhancement. However, the dependency of synchronous multimodal data in the collection and recognition stage brings into the vision weaknesses, such as sensitivity to occlusion and privacy invasion. In this article, we integrate multi-attention structure and auxiliary video information into a novel end-to-end deep learning framework named MuAt-Va, in which video soft labels learned in advance are utilized to teach the multi-attention WiFi feature training process without vision information involved during the test. Specifically, in order to enlarge the application scope and reduce the data cost, we beforehand acquire videos under a satisfactory condition only once, and then leverage teacher–student mechanism to guide the WiFi stream. Instead of straightforwardly concatenating multiantenna channel state information (CSI) from homogeneous wireless signals as previous works, we design a CSI subcarrier-wise, temporal-wise, and view-wise attention module to assign different weights on the basis of data characteristics for the sensing task. Our experiments with multiple subjects data in two scenes demonstrate that MuAt-Va can accurately recognize human actions with more superior performances.