For cybersecurity, privacy protection in big data has received more and more attention and research. Differential privacy is one of the important privacy protection methods, and our work pays attention to differential privacy based on the dynamic taxonomy tree, which can protect the publishing of set-valued data effectively. We propose the optimized dynamic taxonomy tree (ODTT) algorithm as a better and more general way to protect privacy in set-valued datasets. It makes better use of data and reduces noise compared to other privacy-preserving algorithms that use taxonomy tree partitioning. The previous algorithm did not make full use of the characteristics of the dataset when constructing the taxonomy tree, so a 2-itemset’s matrix is used in the proposed algorithm to increase the pseudoempty nodes and reduce the addition of noise. More importantly, we apply the consistency constraint method to the construction of the ODTT algorithm. This retains more statistical characteristics of the original dataset by constraining the noise counts in the leaf partitions of the partition tree. Furthermore, ODTT is extended to deal with dynamic datasets. Finally, we compare the proposed ODTT algorithm with the state-of-the-art CDTT algorithm, by performing a series of experiments and using some evaluation metrics. Experimental results show that ODTT is more general and has higher usability while satisfying the security of differential privacy.

In terms of artificial intelligence, there are several security and privacy deficiencies in the traditional centralized training methods of machine learning models by a server. To address this limitation, federated learning (FL) has been proposed and is known for breaking down "data silos" and protecting the privacy of users. However, FL has not yet gained popularity in the industry, mainly due to its security, privacy, and high cost of communication. For the purpose of advancing the research in this field, building a robust FL system, and realizing the wide application of FL, this paper sorts out the possible attacks and corresponding defenses of the current FL system systematically. Firstly, this paper briefly introduces the basic workflow of FL and related knowledge of attacks and defenses. It reviews a great deal of research about privacy theft and malicious attacks that have been studied in recent years. Most importantly, in view of the current three classification criteria, namely the three stages of machine learning, the three different roles in federated learning, and the CIA (Confidentiality, Integrity, and Availability) guidelines on privacy protection, we divide attack approaches into two categories according to the training stage and the prediction stage in machine learning. Furthermore, we also identify the CIA property violated for each attack method and potential attack role. Various defense mechanisms are then analyzed separately from the level of privacy and security. Finally, we summarize the possible challenges in the application of FL from the aspect of attacks and defenses and discuss the future development direction of FL systems. In this way, the designed FL system has the ability to resist different attacks and is more secure and stable.

With the development of the Internet, a large amount of transaction data (e.g., shopping records, web browsing history), which represents user data, has been generated. By collecting user transaction data and learning specific patterns and association rules from it, service providers can provide better services. However, because of the increasing privacy awareness and the formulation of laws on data protection, collecting data directly from users will raise privacy concerns. The concept of local differential privacy (LDP), which provides strict data privacy protection on the user side and allows effective statistical analysis on the server side, is able to protect user privacy and perform statistics on sensitive issues at the same time. This paper adopts padding-and-sampling-based frequent oracle (PSFO), combined with an interactive query-response method satisfying local differential privacy, to identify frequent itemsets in an efficient and accurate way. Therefore, this paper proposes FIML, an improved algorithm for finding frequent itemsets in the LDP setting of transaction data. The data collector generates frequent candidate sets based on the results of the previous stage and uses them for querying, and users randomize their responses in a reduced domain to achieve local differential privacy. Extensive experiments on real-world and synthetic datasets show that the FIML algorithm can find frequent itemsets more efficiently with the same privacy protection and computational cost.