With the increasing prevalence of outsourcing images to cloud servers, privacy-preserving content-based image retrieval (CBIR) has attracted significant research attention. Existing privacy-preserving CBIR schemes often prioritize retrieval speed by adopting methods that provide weak privacy guarantees and low-dimensional image features, which inevitably compromises security and retrieval accuracy. Additionally, most solutions directly employ CNN models pre-trained on public datasets for feature extraction, neglecting domain adaptation problem. To address these limitations, we propose SecOIR, a secure outsourced image retrieval scheme based on deep hashing networks, which achieves provable security under the semi-honest adversary model while hiding access patterns. Furthermore, domain adaptation is resolved through fine-tuning of feature extraction models. Experimental results demonstrate that SecOIR outperforms state-of-the-art schemes by 11%-12% in accuracy under identical datasets and configurations, while maintaining practical efficiency. To achieve SecOIR, we propose two modified function secret sharing (FSS) schemes that overcome the limited compatibility of the original FSS schemes with replicated secret sharing (RSS). Then, building upon the modified FSS schemes and RSS, we design a series of efficient sub-protocols. Benchmark tests reveal that our sub-protocols surpass existing mainstream solutions in efficiency, which can also serve as independent contributions to secure multi-party computation protocol design.

Skyline query is a fundamental technique in multi-criteria decision-making, aiming to extract “optimal” results that are not dominated by any other data points across all attributes. It has significant value in applications that require trade-offs among multiple criteria. However, existing skyline query methods face two critical limitations: (i) conventional approaches adopt fixed dominance relationships, making it difficult to capture personalized user preferences; and (ii) cloud-based deployment models risk exposing sensitive data and query logic, making it difficult to ensure data privacy and protect query patterns while maintaining efficiency. To address these issues, we propose Privacy-Preserving User-Defined Skyline Query (PUDSQ), a novel privacy-preserving user-defined skyline query framework, which integrates efficient cryptographic techniques–secret sharing (SS) and function secret sharing (FSS)–with a secure database shuffling mechanism to achieve efficient query processing while ensuring robust privacy guarantees. PUDSQ introduces three main innovations: (i) a privacy-preserving filtering framework based on FSS provides dual protection for both data content and user preferences, effectively concealing database content and query logic; (ii) an FSS-based secure protocol suite supporting user-defined attribute retrieval, constrained-region retrieval, and secure skyline filtering; and (iii) a high-dimensional data processing strategy that integrates dimensionality reduction with an Sort-Filter-Skyline (SFS)-based presorting approach to address the high-dimensional data processing challenge and significantly improve efficiency. Experimental results demonstrate that, under equivalent security guarantees, PUDSQ reduces query latency by 8%-90% compared with state-of-the-art solution, with particularly notable advantages in high-dimensional scenarios, achieving an effective efficiency-privacy trade-off.

Cross-modal retrieval is a technique that uses one modality to query another modality in multimedia data (e.g., retrieving images based on text, or retrieving text based on images). It can break down the barriers between different modalities and achieve seamless information connection. Secure cross-modal retrieval focuses on privacy issues in cross-modal retrieval, including private data of data owners and private query requests of users. Current work on secure cross-modal retrieval protects private information through homomorphic encryption, which makes the efficiency of the retrieval phase not ideal. Therefore, the conflict between retrieval efficiency and security has become an important issue that needs to be resolved in secure cross-modal retrieval. We propose a scheme to achieve secure cross-modal retrieval in the form of secret sharing in the IoT environment. In the scheme, the data owner (DO) can secretly divide all the original data into two parts and upload them to two non-collusive cloud servers respectively. The servers store the data and provide cross-modal retrieval for users. The security of the scheme is proved under semi-honest model, and the experiments show that our scheme is more efficient than previous work in the search phase. When the query dimension is 512 and the number of latent factors is 500, the search time is reduced by more than half compared with previous work.

We propose a privacy-preserving machine learning scheme based on the cloud-edge–end architecture to address issues like weak computing power of Internet of Things (IoT) terminals, poor communication quality, and heavy cloud server burdens in traditional frameworks. Edge servers aggregate and forward terminal data, relieving terminals of heavy communication tasks and undertaking part of the computing tasks, which reduces the burden on cloud servers and improves system response speed. For privacy protection, we flexibly use homomorphic encryption and secret sharing techniques, and dynamically add differential privacy noise to resist member inference attacks. Task allocation is coordinated between different layers to optimize computing overhead. Shallow model training is performed on edge servers using homomorphic encryption, while deep model training is conducted on cloud servers using secret sharing. To achieve the conversion from homomorphic ciphertext to secret sharing shares, we design a distributed decryption protocol. Experimental results show our scheme reduces computation overhead by 20%–30% compared to existing privacy-preserving machine learning schemes based on the cloud-edge–end framework, while maintaining privacy protection throughout all stages.

As data continues to grow at an unprecedented rate and informationization accelerates, concerns over data privacy have become more prominent. In image classification tasks, the challenge of insufficient labeled data is common. Transfer learning, an effective and important machine learning method, can address this issue by leveraging knowledge from the source domain to enhance performance in the target domain. However, existing privacy-preserving transfer learning schemes continue to face challenges related to low security and multiple rounds of communication. In the following works, we design a three-party privacy-preserving transfer learning protocol based on the Joint Distributed Adaptation (JDA) algorithm, which ensures malicious security under an honest majority model. To realize this protocol, we designed a series of sub-protocols for constant-round communication, including distributed solving of eigenvalues and eigenvectors based on replicated secret sharing techniques. Compared to existing work, our protocol requires fewer rounds and satisfies malicious security. We provide formal security proofs for the designed protocol and assess its performance using real datasets. Our protocol for computing the eigenvalues of matrices in a given dimension is approximately 2.5 times faster than existing methods. The results of the experiments demonstrate both the security and effectiveness of the proposed approach.

As a practical machine learning method, the K-nearest neighbors (KNN) classification has received widespread attention. The achievement of the KNN classification relies heavily on a large amount of labeled data. However, in the real world, data is often held by different data owners. How to realize efficient joint computing among multiple data owners under the premise of protecting data security and privacy is an urgent problem to be solved. In this paper, we construct a secure multi-party KNN classification scheme (SecKNN) based on function secret sharing (FSS) technology, which is a novel cryptographic primitive and can achieve cheap communication and computation costs for secure computation. Compared with the existing works, our scheme dramatically reduces computational overhead and runs roughly 50.8 times faster than the state-of-the-art approach. Furthermore, our scheme supports the secure KNN classification under general distance functions such as Euclidean distance, Manhattan distance, and Hamming distance. To implement our SecKNN scheme, we design two efficient FSS schemes for Hamming distance function, which implements secure two-party and multi-party Hamming distance computation in a single round. They can be considered as independent research results. Finally, we give formal security proofs for the proposed protocols and validate the effectiveness and efficiency of our protocols through experiments.

With the rapid development of intelligent transportation systems (ITS), more and more intelligent applications for ITS have received widespread attention, such as the vehicle detection, inference of typical routes, and traffic forecasting. In these applications, deep learning is widely used as a key artificial intelligence technology. However, most ITS providers fail to collect enough labeled traffic data for model training. As a complement to deep learning, transfer learning is an effective way to solve the scarcity of labeled data, which can transfer knowledge from labeled datasets to unlabeled datasets, thus improving the accuracy of prediction and classification. Nevertheless, when the labeled dataset and the unlabeled dataset are held by different entities, it is still unrealistic for two mutually distrustful entities to cooperate in transfer learning regarding data security and privacy preservation. Although some existing works provide privacy-preserving transfer learning methods, such methods fail to apply to traffic data with high sample dimensions due to their high computational cost and round complexity. To address this problem, we design an efficient privacy-preserving distributed transfer learning protocol, which is appropriate for traffic data. Compared to existing works, our protocol addresses the privacy-preserving problem of transfer learning for traffic data with high sample dimensions. In addition, our protocol has fewer interaction rounds and can be proved in the semi-honest model. Finally, we validate the effectiveness, efficiency and security of the proposed protocol via experiments. Furthermore, we show the application of the proposed protocol in intelligent transportation systems.

Although secure multi-party computation breaks down data barriers, its utility is reduced when participants have limited computation and communication resources. To make secure multi-party computation more practical, there exists an approach to distribute users' private inputs to multiple servers in a secret sharing manner, and the servers accomplish secure computation tasks through interaction. We propose a new secure computation framework that enables the detection of malicious cloud servers by introducing homomorphic MACs. We utilize pairing-based homomorphic commitments to record MACs on a bulletin board, providing public verifiability while reducing the computation burden on the cloud servers. Additionally, our framework not only supports the underlying general computation, but also prepares for various types of nontrivial high-level operations, such as comparison and bit decomposition. We design a smart payment platform enabling fair payment with the help of smart contracts to protect the rights of both data owners and cloud service providers. Compared to previous works, our framework breaks the limitations of servers being restricted to semi-honest or even honest and provides public verifiability. Performance evaluations demonstrate satisfactory computation and communication efficiency during the online phase of our system.

Summary Difficulties in financing and low utilization of funds are main financial problems that plague the development of small and medium‐sized enterprises. The key to solving this problem lies in opening up the social data circulation between enterprises. It is a good solution for enterprises with frequent data interactions to form groups. Using group loans, the borrowing enterprises could solve the funding difficulties and the loan enterprises could improve the utilization rate of funds. In this article, we construct a group loan system based on blockchain technology, which can promote the free flow of funds among enterprises in the group. We combine the blockchain with the trusted execution environment to realize the automatic determination of loan conditions and realize the automatic execution of smart contracts. We also use the linkable group signature technology to ensure the traceability of loan users while protecting the anonymity. In addition, we use homomorphic encryption technology to make the statement confidential and computable.

In numerous data application scenarios, various data can be represented in the form of data sets, and the intersection is often the common concern of multiple users. Using private set intersection (PSI) protocol, users can securely compute the intersection of their sets without disclosing their private input and other additional information. At the same time, there is also a strong practical demand for statistical analysis of intersection data. However, when multiple parties are involved, the efficiency of the multiparty PSI protocol decreases dramatically as the number of users increases. In this paper, we construct a novel server-aided multiparty PSI protocol, which can transform the complex multiparty computation problem into an efficient two-party computation problem. In addition, we design a series of server-aided party PSI statistical protocols to compute the statistics of the intersection elements, such as the sum, average, variance, range (maximum, minimum), and the cardinality of intersection (the size of intersection). In our protocol, the clients only need to upload their private data to the servers in blinded form and do not need to keep online during server computing. Experiments show that our protocol has high computation and communication efficiency and is suitable for lightweight clients. In addition, we also introduce an application of our protocols in the field of intelligent logistics.

With the rapid development of Internet and the widespread application of distributed computing, people enjoy various conveniences while at the same time their privacy has also been threatened. Secure multiparty computation (MPC) can solve the problem of how data owners who do not trust each other jointly compute in distributed scenarios. Using MPC technique, people can not only realize data joint computing, but also ensure data privacy. In most data application scenarios, private data held by different parties can often be represented by sets. To complete the relevant statistical computations of the intersection of two private sets, we propose a suite of protocols based on MPC. These protocols can compute the statistical functions of the associated data of the intersection, including cardinality, sum, average, variance, range, and so forth, without revealing any additional information other than the result. To achieve these functions, we design a private membership test protocol with the result as the arithmetic sharing value, called the arithmetic shared private membership test (ASPMT) protocol. On the basis of the ASPMT protocol, the size and other statistics of the intersection can be computed securely and efficiently. All fundamental computations are constructed based on secret sharing and oblivious transfer techniques. Thanks to the use of precomputation technique, all protocols are highly efficient.

Convolutional neural network (CNN) is one of the representative models of deep learning, commonly used to analyze visual images. CNN model is more accurate when trained on large amounts of data from multiple sources, and the huge training cost makes the model much more valuable. However, data from various sources is often privacy-sensitive. Therefore, the privacy of these data should be protected during CNN model training and inference. In this paper, we propose an efficient and secure two-party computation (2PC) framework PPCNN for privacy-preserving CNN training and inference. Specifically, we use a new secret sharing technique introduced in ABY2.0 to securely compute various computational tasks involved in the CNN training and inference processes. This secret sharing technique can significantly reduce the communication overhead. Meanwhile, we assign these computationally intensive tasks to cloud servers to reduce the computational burden on local devices. We demonstrate the security of these protocols in the semihonest model. In addition, we use the MP-SPDZ library to simulate our PPCNN framework, and the experiments prove its high efficiency and accuracy.