Clustering algorithms are pivotal tools in data science and machine learning, offering a diverse array of applications ranging from customer segmentation to anomaly detection. With the development of cloud computing and outsourced computing, the adoption of clustering techniques has significantly accelerated. Despite the remarkable benefits of cloud computing, outsourcing sensitive data to remote cloud servers introduces considerable privacy and security concerns. Specifically, there is a risk that cloud service providers may engage in malicious behavior, such as data tampering. In response to these concerns, we propose four basic protocols based on vector space secret sharing, including secure Euclidean distance protocol, comparison protocol, minimum protocol, and division protocol. By applying these protocols, we construct a secure clustering scheme that can identify malicious behaviors. We thoroughly demonstrate the security of each underlying protocol as well as the overall clustering scheme. To validate the practicality and effectiveness of our approach, we conduct experiments on standard datasets. The results show that our clustering scheme performs efficiently while maintaining strong security guarantees.

The development of cloud computing needs to continuously improve and perfect the privacy-preserving techniques for the user’s confidential data. Multi-user join query, as an important method of data sharing, allows multiple legitimate data users to perform join query over the data owner’s encrypted database. However, some existing join query protocols may face some challenges in the practical application, such as practicality, security, and efficiency. In this article, we put forward a dynamic and secure join query protocol in the multi-user environment. Compared with some existing protocols, the proposed protocol has the following advantages. On the one hand, we utilize the dynamic oblivious cross tags structure to realize an efficient join query with forward and backward security. On the other hand, we combine the randomizable distributed key-homomorphic pseudo-random functions with join query to support multiple data users, which can provide resilience against the single user’s key leakage and resist collusion attacks between the cloud server and a subset of data users. We formally define and prove the security of proposed protocol. In addition, we give a detailed analysis of computation and communication overheads to demonstrate the efficiency of proposed protocol. Finally, we carry out some experimental evaluations to further demonstrate the superiority of functionality and efficiency.

Generative Pretrained Transformer (GPT) is an advanced natural language processing (NLP) model and is excellent at understanding and generating human language. As GPT is increasingly utilized, more and more cloud inference services for pre-trained generative models are being offered. However, when users upload their data to cloud servers to experience cloud inference services, ensuring the privacy and security of their data becomes a challenge. Thus, in this work, we present SecureGPT, a framework for multi-party privacy-preserving transformer inference in GPT and design a series of building blocks which include M2A (conversion of multiplicative share to additive share), truncation, division, softmax and GELU protocols for our framework. Specifically, we follow the work of SecureNLP and further explore the M2A protocol for non-linear functions such as GELU and softmax. We also design multi-party private protocols for GPT’s transformer sub-layers. Finally we prove the security of our framework in the semi-honest adversary model with all-but-one corruptions. we evaluate the runtime of our framework under different parties settings and our implementation leads to up to$100\times $improvement compared to state-of-the-art works.