Fully homomorphic encryption (FHE) is a powerful technique that allows unlimited computations on encrypted data without decryption. However, FHE will incur huge computation and storage costs, making it difficult to be applied in real environments. To improve the efficiency of FHE, some optimized schemes have been proposed based on the trusted execution environment (TEE), which offer a lighter and lower overhead solution for FHE optimizations to a certain extent. However, they heavily rely on the confidentiality of the TEE, and their performance is still limited. To solve the above problems, we propose an efficient TEE-assistance CKKS scheme without bootstrapping, named TCKKS, which has the characteristics of security, efficiency, low memory, and scalability. First, to weaken the trust assumption of TEE, we consider TEE to be honest-but-curious, meaning the enclave's algorithm provider will execute the algorithm honestly but might monitor the data in the enclave. Based on this assumption, we design a lightweight secure multiplication protocol (SMP) and a secure rotation protocol (SRP) for TCKKS to efficiently achieve ciphertext multiplication and rotation operations. Then, to further improve the performance of TCKKS, we optimize the arithmetic operations and encryption/decryption operations based on the characteristics of our protocols. Moreover, we prove the security of SMP and SRP under the simulation-based real/ideal worlds model and further demonstrate the security of TCKKS based on the RLWE problem. In addition, extensive experiments indicate that TCKKS has a better performance than mainstream libraries, such as RNS-HEAAN, PALISADE and SEAL.

Promoting data sharing is one of the critical strategies for thriving in the digital age, and enormous demand for location-based services propels data sharing in the Internet of Vehicles (IoV), particularly in the case of integrating mobile crowdsensing (MCS). However, data security and privacy concerns are increasingly posing serious challenges to the development of data sharing. Although a slew of works have been designed to achieve secure data sharing in IoV, they are inadequate for addressing the privacy issues identified in the data sharing model and often suffer from performance limitations. In this work, we propose an efficient and secure data sharing scheme under the MCS-integrated IoV. Specifically, motivated by the distributed point function (DPF), we design a double-output DPF and leverage it to construct a secure updating scheme that protects full privacy while ensuring high efficiency. Then, based on the XOR filter and a series of subtle transformations, we carefully design a secure spatial test protocol to determine whether a point falls within an arbitrary spatial range efficiently. Afterward, we propose a secure retrieving protocol by using the idea of shared shuffling, in which the offline sub-protocol is presented to generate random masks, and the online sub-protocol is designed to quickly retrieve the desired data items. After formally proving the security of our proposed schemes, we experimentally evaluate their efficiency by comparing them with the alternative solutions, and the results indicate that our proposed schemes offer superior performance, particularly in terms of communication overheads.

In the advancing field of information technology, the Industrial Internet of Things (IIoT) plays a pivotal role across various domains. To fully leverage the massive data generated by IIoT, the Balanced Iterative Reducing and Clustering using Hierarchies (BIRCH) algorithm is well-suited for clustering large-scale time series data due to its incremental and order-sensitive design. Although aggregating multisource data on cloud servers becomes a practical and effective approach, it raises privacy concerns. Existing privacy-preserving BIRCH schemes designed to mitigate these concerns, however, rely on homomorphic encryption for atomic operations and reveal the insertion path of sample points, resulting in significant limitations in performance and security. To address these issues, we propose SBIRCH, a novel secure BIRCH scheme. Specifically, we first introduce oblivious hierarchical updates to preserve insertion paths. Then, we present a shuffling-based obfuscation technique to protect intermediate clustering results during node splitting. In addition, we optimize division operations on encrypted data, delivering high precision and significantly improving efficiency. Security analysis confirms that our scheme effectively protects multisource data, intermediate clustering results, and clustering relationships. While maintaining consistent security, experimental evaluations demonstrate that our scheme achieves substantial improvements in computational efficiency and markedly reduces communication overheads compared to existing schemes.

Merging multi-source time series data in cloud servers significantly enhances the effectiveness of analyses. However, privacy concerns are hindering time series analytics in the cloud. Responsively, numerous secure time series analytics schemes have been designed to address privacy concerns. Unfortunately, existing schemes suffer from severe performance issues, making them impractical for real-world applications. In this work, we propose novel secure time series analytics schemes that break through the performance bottleneck by substantially improving both communication and computational efficiency without compromising security. To attain this, we open up a new technique roadmap that leverages the idea of mixed model. Specifically, we design a non-interactive secure Euclidean distance protocol by tailoring homomorphic secret sharing to suit subtractive secret sharing. Additionally, we devise a different approach to securely compute the minimum of three elements, simultaneously reducing computational and communication costs. Moreover, we delicately introduce a rotation concept, design a rotation-based hybrid comparison mode, and finally propose our fast secure top-$k$protocol that can dramatically reduce comparison complexity. With the above secure protocols, we propose a practical secure time series analytics scheme with exceptional performance and a security-enhanced scheme that considers stronger adversaries. Formal security analyses demonstrate that our proposed schemes can achieve the desired security requirements, while the comprehensive experimental evaluations illustrate that our schemes outperform the state-of-the-art scheme in both computation and communication.

With the rapid advancement and widespread ap plication of the graph neural network (GNN), the collaborative graph learning, in which multiple parties collaboratively construct a GNN model using their respective graph data, has attracted increasing attention. However, this paradigm also raises significant privacy concerns, as both nodes and edges may contain sensitive personal information, while existing privacy preserving schemes often come at the cost of degraded model performance or substantial system overhead. Therefore, this paper proposes an efficient and privacy-preserving collaborative, and Hui Li, Member, IEEE, Xiaoyu Kou Social Platform learning framework on vertically partitioned graph data, dubbed Plog. Specifically, we first design a decomposition algorithm to split the sparse adjacency matrix into the summation of multiple independent permutations, which are lightweight, parallelizable, and well-suited for secure multi-party computation. Building on this, a weighted oblivious batch permutation protocol is carefully customized based on correlated randomness to securely and efficiently compute adjacency matrix multiplications, addressing the core efficiency bottleneck in GNN inference and training. The selective security of Plog is formally verified under the ideal-real paradigm. Extensive experimental results on three real world datasets demonstrate that compared to the state-of-the art scheme, Plog can reduce online communication rounds by 46% and achieve a 1.73× speedup in the overall inference and training time.

Fully Homomorphic Encryption (FHE) allows computations on encrypted data. However, FHE faces significant performance overhead due to the time-consuming homomorphic multiplications and bootstrapping operations, which limits its practical applications. Meanwhile, many FHE optimization schemes have been proposed based on a Trusted Execution Environment (TEE), but they heavily rely on the trustworthiness of the TEE. In outsourcing scenarios, the TEE owner might be a third-party service provider, posing a risk of data leakage for users. To address these issues, we propose an efficient BGV scheme with semi-honest TEE assistance. We consider the TEE to be honest-but-curious, meaning the TEE owner might eavesdrop on the user's data during computation. Based on this assumption, we design a new BGV homomorphic multiplication protocol that operates in a smaller ciphertext space without compromising the original security level and substantially improves computational efficiency. With the assistance of a semi-honest TEE, our multiplication protocol eliminates the need for the switch key, reducing initialization time and supporting an unlimited number of homomorphic multiplication operations without bootstrapping. Furthermore, we provide a security proof using a simulationbased real/ideal world model. Extensive experiments show that our scheme outperforms mainstream libraries such as SEAL, OpenFHE, and Helib.

Machine Learning (ML) models have proven effective in optimizing wireless and private networks. However, recent research highlights the threat of data poisoning attacks on ML models. To analyze such a threat on an industrial 5G private network, this work investigates the effectiveness of data poisoning against it. We primarily focus on poisoning four ML models: Support Vector Machines (SVM), Random Forest (RF), Decision Tree (DT), and Artificial Neural Networks (ANN), at three poisoning levels: 10%, 15%, and 20%. Our research shows that all models introduce instability in the network, rather than optimization, whereas neural networks are less affected by data poisoning compared to other models. At the 20 % data poisoning level, model performance degrades by about 6-7% for SVM, RF, and DT, while ANN shows a minimal disruption of 2%.

Weighted set sampling has been proven essential for generating discrete numbers based on their weights and found broad applications in recommendation systems. The extension of this method, known as weighted range set sampling (WRSS), specifies a query range and applies weighted set sampling to the data within that range. With the proliferation of cloud computing, outsourcing encrypted data and data processing tasks to cloud servers has become a common practice to overcome data storage and processing challenges while protecting data privacy. Existing studies have proposed many privacy-preserving solutions for various customized query and data processing tasks, none have specifically addressed privacy-preserving WRSS. In response to this gap, our paper introduces an efficient and privacy-preserving WRSS scheme. We begin by leveraging the three-party secret sharing (TPSS) scheme as a foundation to design an enhanced three-party secret sharing (eTPSS) scheme with superior storage and computational efficiency. Building upon the eTPSS scheme, we introduce a series of private algorithms to safeguard WRSS privacy. Our scheme integrates the use of a binary search tree and the alias method for WRSS, ensuring privacy through eTPSS-based private algorithms. A thorough security analysis under the simulation-based real/ideal worlds model showcases the effectiveness of our proposed scheme. The proposed scheme's efficiency has been substantiated through extensive experiments, demonstrating that our scheme marks a significant advancement in addressing the challenges posed by privacy-preserving WRSS.

Geophysical time series derived from remote sensing and ground-based observations contain rich signals reflecting diverse Earth processes. Wavelet filtering is widely employed to extract these signals from noisy datasets. However, due to many factors, geophysical time series inevitably contain missing data, complicating the direct use of wavelet filtering without prior interpolation. The extended wavelet filtering (EWF) method (Ji et al. [1]) addresses this by linking residuals (signals) to missing data and estimating signals by minimizing the quadratic norm of residuals across both observed epochs and missing epochs. However, EWF is suboptimal for filtering observed data, as residual estimates for missing epochs are less reliable, particularly in time series with consecutive data gaps. This study proposes a new approach, minimum norm least squares wavelet filtering (MWF), for processing gappy time series. MWF offers two main advantages over EWF: (a) it filters incomplete time series more effectively through optimal temporal filtering by minimizing the quadratic norm of residuals only for observed epochs. This can avoid disturbances in filtering caused by distorted residuals at missing epochs in EWF. To address the rank deficiency in the parametric model, an additional constraint is applied to minimize the L2 norm of unknown parameters (missing data), ensuring a unique solution. (b) it provides a more concise and effective way to consider prior precision information of the time series by incorporating it into the cost function, without the need for iterative normalization required by the EWF method. We apply the proposed method to extract deformation signals from daily position time series of 27 Global Navigation Satellite System (GNSS) permanent stations in the Chinese mainland and compare the results with the EWF method. Results show that MWF outperforms EWF in signal extraction, as indicated by smaller fitting errors and higher signal-to-noise ratios. Additional simulations confirm that MWF yields signals closer to the true signals regardless of gap sizes, noise characteristics (pure white or colored noise), and gap distributions (uniform or consecutive).

Vertical federated learning for boosting trees has gained significant attention due to its ability to enable participants to collaboratively train high-quality models while preserving data privacy. However, existing privacy-preserving vertical boosting tree schemes suffer from high computation and communication costs or potential security vulnerabilities. Recently, SGBoost, a federated outsourced training and inference scheme, was proposed to address these challenges. However, its performance and security still require significant improvements. Therefore, we propose SGBoost+, an efficient and privacy-preserving vertical boosting tree framework for federated outsourced training and inference. Building upon the strengths of SGBoost, we introduce an RLWE-based lossless and secure internal node construction and an efficient oblivious inference algorithm to finish the model training and inference, significantly enhancing both security and efficiency. To reduce communication cost, we design a ciphertext compression algorithm for model training, which drastically minimizes data transmission costs. Additionally, we analyze the security of a symmetric encryption scheme, specify the required security conditions and parameters, and optimize our model inference based on its improved and secure version. Detailed security analysis confirms that SGBoost+offers strong privacy guarantees. Extensive experiments demonstrate that SGBoost+achieves efficient model training and inference with significantly lower computation and communication costs compared to state-of-the-art schemes.

The rapid development of big data and Internet of Things has promoted the formation of data silos, and cloud computing has facilitated the outsourcing of vertically distributed data to cloud servers. In outsourced query scenarios, building query indexes is crucial for balancing data utility and data privacy protection. The hypercube tree is a widely used index for multi-dimensional data, supporting various query types. Although secure hypercube tree-based queries have been extensively studied in existing works, they are not applicable for building a hypercube tree over vertically distributed ciphertext data. To address this issue, we propose the first efficient and secure hypercube tree building scheme for vertically distributed data, named SCTBuild. We first design a flexible three-party secret sharing (fTPSS) scheme, allowing data owners to flexibly configure secret sharing forms based on real-world computational, communication, and storage constraints. Then, we design a communication-efficient data outsourcing algorithm, a secure data permutation algorithm, and a secure data comparison algorithm based on the fTPSS scheme. After that, we propose our SCTBuild scheme based on the aforementioned algorithms, in which data owners first perform pre-computation on their data to improve tree-building efficiency. We prove that our fTPSS scheme, private algorithms, and the SCTBuild scheme are semantically secure in the simulation-based real and ideal worlds security model; and conduct experiments to validate their high efficiency.

Logistic regression is a widely used and efficient machine learning algorithm for data analysis. However, data is often distributed among different participants, and building high-quality models often involves data interaction, which might easily lead to sensitive information leakage. While some schemes have been proposed to address the data security issue through privacy-preserving technologies such as multi-party computation, homomorphic encryption, and trusted execution environment (TEE), they still suffer from some limitations in terms of security, efficiency, and model integrity and are not suitable for the secure outsourcing training. Therefore, in this paper, we propose an efficient and lossless integrity-preserving training for high-dimensional logistic regression over vertically partitioned data. First, we utilize a block strategy to encrypt the data in batches and outsource them to a cloud with TEE to achieve the non-interactive training. Considering TEE’s limited memory, we employ random sampling technology to achieve mini-batch training within TEE, which ensures the data security. We find that the cloud might launch the active attacks to drop the model accuracy, such as lazy attack and replacement attack. To prevent the attacks, we introduce an integrity protection mechanism to verify data integrity and reduce verification frequency using a mixed-probability checking method, including fixed and random probability checking. Detailed analysis has confirmed that our scheme is secure and integrity-preserving. Extensive experiments have demonstrated that our scheme can achieve lossless, non-interactive, and efficient model training.

Big data have witnessed a growing trend towards vertically distributed storage, with various queries on vertically organized data recognized as effective means for unlocking data's inherent value. Several solutions have emerged for enabling privacy-preserving queries on vertically distributed data using secure multi-party computation techniques. However, these approaches often involve substantial communication overheads among data owners and place significant computational burdens on them, rendering them impractical for resource-constrained data owners. Outsourcing vertically distributed queries to the cloud can substantially alleviate the computational burdens on data owners, and efficient index construction is crucial for outsourced queries on vertical data. In light of this, we present the pioneering “Privacy-Preserving k-d Tree Building” (PTreeB) scheme for vertically distributed outsourced data in this study. Our scheme begins with the development of a private random dimension choosing algorithm (PCDim) and a private equality test (PET) algorithm, leveraging additive Paillier homomorphic encryption. Subsequently, these algorithms, along with various efficiency-enhancing strategies, including pre-sorting each data owner's data and adopting a dual-key system for data privacy protection, form the foundation of our PTreeB scheme. We rigorously demonstrate the security of our scheme, and its efficiency is validated through extensive experimentation.

Machine learning-based trajectory-user linking (TUL) task, which explores human mobility patterns and identifies known users from anonymous trajectories, is widely applied in location-based personalized recommendation systems. However, TUL models implicitly and inadvertently retain users historical trajectories in training data, which could be revealed under malicious inference and analysis. To address this issue, we propose UNIRE, a knowledge transfer-style TUL model aggregation framework to protect sensitive training data. Specifically, teacher TUL models are trained with private users trajectories and then utilized to collaboratively annotate a sensitive-trajectories-isolated dataset. Leveraging this teacher-labeled dataset, student TUL model can be trained and released as the secure aggregation of teacher models, preventing attackers from arbitrarily accessing or exposing the private trajectories. Meanwhile, differential privacy is applied for safeguarding the knowledge transfer process, ensuring that aggregation information from individual teacher models is indistinguishable. Furthermore, we introduce trajectory Equalization and Pseudo Classification as data alignment and accuracy optimization mechanisms for training and aggregating, respectively. Finally, security analysis and extensive experiments indicate that UNIRE framework achieves effective sensitive training data preservation and performance-improved TUL models aggregation, compared to state-of-the-art baselines.

XGBoost, a widely-used machine learning algorithm, has been applied across various fields. To develop a high-quality model, federated learning is employed, allowing participants to keep their data localized while sharing gradients for global model training. However, recent studies have shown that gradients can leak sensitive information, and even when protected, they might still be vulnerable to certain attacks. While several privacy-preserving and verifiable schemes have been proposed to address this issue, they still face limitations in terms of computational and communication costs in certain scenarios. To solve the problem, we propose a privacy-preserving and verifiable XGBoost training scheme for horizontal federated learning. First, we propose a masking with one-time padding protocol (MOTP) to securely aggregate gradients, supporting both online and offline modes for generating random masks. Both approaches reduce the communication costs and improve the training efficiency. Next, we present a linear aggregation verification protocol (LAVP) to ensure the gradient integrity, which avoids complex computations, thereby improving verification efficiency. Building on MOTP and LAVP, we propose a privacy-preserving and verifiable internal node construction algorithm to train a XGBoost model for horizontal data. Detailed security analysis and extensive experiments demonstrate that our scheme can achieve the privacy-preserving, verifiable, lossless, and efficient XGBoost training for horizontally partitioned data.

Understanding terrestrial water storage (TWS) changes is crucial for water management and hydrological applications. TWS changes are accurately observed by the Gravity Recovery and Climate Experiment and its Follow-On (GRACE/-FO) mission. However, the low spatial resolution limits the knowledge of water storage distribution. This study proposes a novel constrained point-mass modeling (CPM) approach, introducing a data-driven regularization matrix to improve spatial resolution. We derived point-mass solutions over the Amazon River basin from April 2002 to December 2019. Then, we evaluated our method using the WaterGAP global hydrology model (WGHM). Compared to results from traditional point-mass method (TPM) and three state-of-the-art GRACE/-FO solutions, the annual amplitudes in TWS changes from our method agree better with that from WGHM (slope =0.80 and$R^{2} =0.69$). Besides, the 179-month TWS changes also show a better consistency between our results and WGHM, indicating that our method effectively improves spatial resolution over the Amazon River basin. Moreover, benefiting from the improved spatial resolution of our method, our results, based solely on GRACE/-FO data, reveal spatial patterns in TWS changes that generally correspond with the major river channels of the basin.

Big data and bursting cloud computing technologies have facilitated an increasing trend of outsourcing data-driven services to the cloud, where the reverse kNN (RkNN) query is a popularly outsourced query service. The RkNN query aims to retrieve objects having the query object as kNN and widely applied in the product recommendation. Considering privacy concerns, the outsourced query services are demanded to protect data privacy, and consequently a series of privacy-preserving query solutions have been put forth. Nevertheless, RkNN query over high-dimensional data has not been studied to date. In this work, we design the first efficient and privacy-preserving RkNN query scheme over encrypted high-dimensional data, named PHRkNN. Specifically, we first introduce a pivot filter condition for the RkNN query and utilize it to deliberately design a pivot filter R-tree (PFR-tree) to organize the high-dimensional dataset such that the RkNN query has sublinear query efficiency. Then, we propose our PHRkNN scheme by designing some homomorphic encryption based private algorithms and applying them to privately achieve PFR-tree based RkNN query. After that, we propose an oblivious PHRkNN scheme on the basis of the PHRkNN scheme by designing a private random tree permutation (PRTP) algorithm to protect the access pattern privacy. The security of our PHRkNN scheme and oblivious PHRkNN scheme is proved by the simulation-based security analysis. The performance is verified through computational costs and communication overheads evaluation.

Wavelet multiresolution analysis (MRA) is a widely used and effective method for filtering noisy time series and investigating the detailed characteristics of time series at different resolutions. However, geophysical time series often exhibits unavoidable data gaps stemming from diverse factors, impeding the straightforward implementation of ordinary MRA (OMRA). Moreover, geophysical time series is typically heterogeneous, as evidenced by the dynamic fluctuations in their precision. Regrettably, the prevailing approach of OMRA neglects to incorporate this aspect of heterogeneity. This study develops an extended MRA (EMRA) approach that solves for the missing values based on the best approximation in the temporal domain. The proposed method can directly analyze incomplete time series without prior interpolation and accounts for the formal errors of time series to improve the filtering performance. To validate the proposed approach, we use the EMRA method to extract crustal deformation signals from the daily position time series of 27 permanent global navigation satellite system (GNSS) stations in the Chinese mainland from 1999 to 2019. We compare the results with those obtained through OMRA using interpolation methods. The results reveal that EMRA can extract more signals than OMRA, particularly when considering formal errors. Repeated simulations further depict that the signals extracted by EMRA are closer to the simulated true signals than those by OMRA using interpolation methods.

With the rapid advancements in machine learning and the widespread adoption of Model-as-a-Service (MaaS) platforms, there has been significant attention on convolutional neural network (CNN) inference services. However, traditional inference services over plaintext data and models are susceptible to the risks of data and model leakage. Although several privacy-preserving CNN inference schemes utilizing trusted execution environment (TEE) and cryptography have been proposed, their security models and performance still have limitations in some scenarios. Aiming at the above challenges, we present an oblivious neural network prediction scheme with semi-honest TEE, namely ToNN, which ensures the security of users’ inputs, outputs, and the model itself. Specifically, based on the limited memory of the TEE, we design secure protocols to perform CNN calculations securely and efficiently, which are friendly to support the single instruction multiple data technique. Additionally, we propose a look-up-table method to optimize the convolution and pooling layers calculations. A detailed security analysis under the simulation-based real/ideal worlds model shows that ToNN can achieve the desired security. Extensive simulation results further demonstrate that ToNN can improve the performance of linear calculations by$\textbf {4.86}\times $and non-linear calculation by$\textbf {37.68}\times $, and can be implemented effectively with low computation and communication costs.

The increasing prevalence of cloud computing drives the exploration of various secure query schemes over encrypted data, among which secure spatial keyword query has drawn a great deal of attention due to its broad application in location-based services. However, most existing schemes are either limited to the boolean keyword test or incapable of protecting access pattern privacy. Although the state-of-the-art secure spatial keyword query scheme can support keyword similarity while preserving access pattern privacy, it is unable to cope with the arbitrary spatial range, which is more general, and has limitations in efficiency and security. In this paper, we propose a new secure spatial keyword similarity query scheme that can support arbitrary spatial ranges and enhance the efficiency and security of the state-of-the-art scheme at the same time. Specifically, we first present a new homomorphic encryption technique by improving the popular symmetric homomorphic encryption (SHE). After that, we propose a novel approach to make supporting arbitrary spatial ranges over encrypted data possible, in which a spatial encoding technique is designed to improve performance. Finally, by designing a pack-based solution to protect access pattern privacy, our proposed scheme can hide the number of query results while optimizing performance. We formally prove the security of our proposed scheme and conduct experiments to evaluate its performance. The results indicate that our proposed scheme outperforms the state-of-the-art scheme in both the computational costs and communication overhead.

Outsourcing big data to cloud servers has gained prominence, and growing concerns about privacy, alongside privacy-related regulations, underscore the need to encrypt data before sending them to the cloud. Nevertheless, encryption significantly hampers the query capabilities of data, particularly in the case of vertically distributed data. This paper focuses on developing secure and efficient similarity query schemes for vertically distributed data in cloud environments. As is known, current solutions are constrained by limitations in query efficiency, approximate query results, and their ability to support vertical data. To address these issues, we introduce two novel schemes: a Fast Similarity Query Scheme (FSQ) and a Non-interactive Similarity Query Scheme (NoSQ) for outsourced distributed data. In the FSQ scheme, we enhance query efficiency by designing a trusted execution environment (TEE) assisted fast secret sharing (FSS) scheme and a series of FSS-based private algorithms, enabling secure data index construction and fast similarity query processing. For the NoSQ scheme, we eliminate communication overheads by designing a TEE assisted non-interactive secret sharing (NoSS) scheme and a series of NoSS-based private algorithms. Both schemes have undergone rigorous security validation using a simulation-based real/ideal worlds model, and their efficiency has been confirmed through comprehensive experiments.

With the rapid growth of Internet data volumes, Big Data analysis technologies have gradually permeated all aspects of life. However, the existence of data silos and the promulgation of relevant regulations make it challenging to apply these technologies. In this context, federated learning provides a feasible solution. Especially, XGBoost schemes for vertical federated learning have attracted much attention due to the widespread use of XGBoost. However, these schemes have limitations in terms of security or efficiency. To address these issues, we propose an efficient and privacy-preserving vertical federated learning framework based on the XGBoost algorithm, namely ELXGB, which achieves secure data alignment, XGboost training, and inference services. First, we design two node split algorithms based on homomorphic encryption and differential privacy, which securely and efficiently achieve tree node generation to construct the global model. Then, we utilize attribute obfuscation and direction obfuscation to achieve a secure inference algorithm, which avoids sensitive information leakage and protects the global model. Additionally, the global model of ELXGB is designed to be centralized, which does not require all participants to stay online for inference. Detailed security analysis demonstrates that ELXGB is privacy-preserving. Moreover, extensive experiments on real-world datasets indicate that ELXGB achieves high efficiency without sacrificing model accuracy.

The ever-growing data scale and increasingly strict privacy restraint have recently drawn extensive attention to federated learning (FL) as a multi-party machine learning paradigm for achieving high-quality model construction without data collection. Nevertheless, uploading local models in FL can still be exploited by adversaries to infer participants' sensitive data. Furthermore, it is possible for malicious participants to manipulate the global model by submitting poisonous local models. To tackle these challenges, this paper proposes an efficient and privacy-preserving federated learning framework against poisoning adversaries, namely ELFL, which can ensure the confidentiality of local models while effectively resisting data poisoning attacks. Specifically, we first design a grouped secure aggregation algorithm, through which the aggregation server can compute the summations of local models inside logic groups but cannot see individual ones. Then, based on grouped aggregations, our poisoning defense mechanism could detect and quickly phase out malicious participants from training candidates. Moreover, the computational complexity of participants is independent of their total number, so it is suitable for large-scale scenes. Detailed security analysis demonstrates the security of ELFL. Experimental results show that ELFL could maintain a high accuracy against representative data poisoning attacks, and its computational and communication overhead is indeed low.

Set similarity query is a fundamental query type in various applications, such as clinical diagnosis, online shopping, and mobile crowdsensing. Meanwhile, as the prevalence of outsourced query services, privacy-preserving set similarity query has been considerablely studied. However, to the best of our knowledge, most previously reported solutions suffer from applicability, efficiency, or security issues. Aiming at addressing these issues, we propose an efficient and privacy-preserving set similarity range query scheme (EPSet), where Jaccard similarity is employed as the similarity metric. Specifically, the set similarity range query is first transformed into multi-dimensional range queries by leveraging the triangle inequality of Jaccard distance. Then, a pivot-based k-d tree is designed for indexing the dataset and processing the set similarity query. After that, we design homomorphic encryption based privacy-preserving filter/refinement protocols, respectively named as PPF and PPR, to protect set similarity query privacy, and propose our EPSet scheme. The security of our scheme is proved under the simulation-based real/ideal model, and the performance is validated thorugh the extensive experiment evaluation.

The explosive growth of data scales has recently drawn extensive attention to machine learning (ML) prediction services for achieving effective and efficient decision-making. However, the potential risk of model or data leakage, as well as vertically partitioned query data in real-world scenes, bring many challenges to ML prediction. Therefore, this paper proposes an efficient and privacy-preserving logistic regression prediction scheme, through which query institutions with vertically partitioned data can enjoy secure, fast, and high-accuracy prediction services from one cloud service provider. Specifically, we take the efficiency advantage of the trusted execution environments (TEE) technique under a semi-honest model, and design a series of masking and recovery methods based on homomorphic encryption to guarantee privacy inside and outside the enclave. Moreover, the ciphertext packing technique is integrated into our scheme for processing predictions parallelly, which further reduces communication and computational overhead. As a result, our scheme will not introduce additional strong security assumptions to TEE; meanwhile, it achieves a trade-off between security and efficiency. Based on the simulation-based paradigm, we formally prove the selective security of our scheme. Extensive experiments on real-world and synthetic datasets show that our prediction accuracy is comparable with plaintext prediction, and our scheme has at least a 30× running speedup compared to the existing representative schemes.

The hidden Markov model (HMM) has played a pivotal role in various IoT applications due to its ability to model time-varying sequences. Since the datasets usually live in isolated islands and their privacy naturally demands to be seriously considered, the HMM should be trained in a privacy-preserving manner. A typical HMM training framework is federated learning, in which a federated server and many data owners collaboratively train an HMM without revealing data owners' data to the federated server and the trained model to data owners. Since existing HMM training schemes are computationally intensive, we propose an efficient and privacy-preserving federated learning scheme for HMM training to address the efficiency issue in this paper. First, we transform all HMM training computations into matrices- and vectors-based computations over real domains. Then, we introduce our federated HMM training scheme by applying matrix encryption to protect the HMM training privacy. After that, we show that our scheme is privacy-preserving through a rigorous analysis on the security of our scheme. We illustrate that our scheme is efficient through extensive experimental evaluation on the performance of our scheme.

With the rapid development of machine learning, MLaaS has infiltrated into many fields such as image recognition, natural language processing, medical diagnosis, and so on. However, in MLaaS, data interaction between users and service providers is inevitable, and both users' private data and servers' model parameters are at risk of privacy disclosure. In order to solve this problem, homomorphic encryption is an extensively used technique to process private information over ciphertexts. However, since homomorphic encryption only supports linear operations, approximation techniques are required to calculate nonlinear functions, which leads to the loss of prediction accuracy and heavy computation overhead. Therefore, in this paper, we propose a secure neural network prediction scheme combining the trusted execution environment and homomorphic encryption with different security assumptions. Specifically, we first define the security model of TEE-assisted neural network prediction. Then, by combining a lightweight homomorphic encryption technique with TEE, we design secure neural network prediction protocols under different security levels, with which neural network prediction can be securely processed with high performance and accuracy. Finally, we evaluate the performance of our scheme on the MNIST, Fashion-MNIST, and KMNIST datasets, and the results demonstrate that our scheme indeed improves the prediction efficiency and accuracy compared to traditional homomorphic encryption-based schemes with polynomial approximation.

With the development of location based service and online social networking, geo-social-based points of interest (POIs) recommendation has received wide attention, which comprehensively considers the geographic and social factors. The popularity of cloud computing techniques have driven the emerging trend of outsourcing the geo-social-based POI recommendation service to the cloud. However, the cloud server is not fully trusted, leading to the raising concerns of data privacy. Although many privacy-preserving schemes have been proposed for the geo-social-based POI recommendation, they can only return approximate query results. Aiming at addressing this issue, in this paper, we propose an efficient and privacy-preserving geo-social-based POI recommendation scheme, called TRIPE, with accurate query results. Specifically, we first leverage the Quadtree to organize the geographic data and the MinHash method to index the social data. Then, we design a Quadtree-based POI filtering algorithm and a MinHash-based POI filtering algorithm to filter out some POIs that do not meet geo-social POI recommendation threshold. Meanwhile, we employ the BGV homomorphic encryption to protect the privacy of Quadtree-based/MinHash-based POI filtering algorithms and propose our TRIPE scheme based on these algorithms. Security analysis shows that TRIPE is privacy-preserving, and experimental results show that TRIPE is efficient.

Cloud computing's ease of management and on-demand availability features have triggered the boom of outsourcing query services to the cloud. Considering the not-fully trustfulness of the cloud server, the outsourced query services should take the data privacy into consideration, especially when sensitive data are involved. Although many query types have been studied in the context of cloud computing, the range-based reverse k nearest neighbors (RkNN) query is still an unexplored area, which searches data records having any record in the query range as kNN and has wide applications in the taxi dispatching and advertisement placement. As a steppingstone, we propose the first efficient and privacy-preserving range-based RkNN query (PRRkNN) scheme in the cloud. We first design a modified R-tree (MR-tree) to organize the dataset and introduce an efficient MR-tree based RkNN query algorithm to handle RkNN queries with sublinear query efficiency. Then, we propose a sign-preserving matrix encryption (SPME) scheme to privately determine the sign of the scalar product between two records and deploy Paillier homomorphic encryption to introduce a two-party multiplication protocol for secure multiplication between a matrix and a vector. After that, we leverage the SPME scheme and the multiplication protocol to design our PRRkNN scheme. In addition, security analysis shows that our PRRkNN scheme achieves the desired security; and the performance evaluation verifies its efficiency.

With the ubiquity of Internet of Things, location-based service (LBS) providers have collected huge volumes of individuals’ trajectories, which are valuable for some applications, e.g., store location choosing for merchants. However, directly publishing raw trajectories to applications may violate individuals’ data privacy and lead to unexpected loss. Although many trajectory synthesis methods under differential privacy have been proposed to privately publish trajectories data, they cannot sufficiently preserve the semantic information of trajectories. Aiming at this issue, in this article, we introduce a semantic-preserving scheme to synthesize trajectories for publishing under differential privacy. Specifically, we first design a hierarchical graphical model (HGM) to capture the semantic feature of trajectories. Then, we propose a metric, named the correlation score, to measure the relationship between two locations, which can well capture the geographic feature of trajectories. After that, we propose a private trajectory synthesis algorithm by first adding Laplace noises to the extracted features and then synthesizing trajectories based on the noisy features and the Markov chain theory. Privacy analysis demonstrates that our scheme can protect the privacy of trajectories. In addition, performance evaluation illustrates that our synthetic trajectories maintain good utility semantically and geographically.

With the explosive growth of data volume and computing capability, federated learning, which involves constructing global models over multiple data islands, has demonstrated its advantages and vast prospects in the field of machine learning. However, due to commonly vertically partitioned data, coupled with privacy concerns about data leakage, there are still some challenging issues in traditional federated learning. To tackle these challenges, in this article, we propose an efficient and privacy-preserving vertical federated learning framework for logistic regression, named VFLR, where multiple participants can collaboratively perform global model training and query over their vertically partitioned data. Specifically, we first design a data aggregation matrix construction algorithm, with which the vertically partitioned data can be aggregated for high-accuracy global model training. Then, by utilizing a novel symmetric homomorphic encryption, our framework can ensure that the whole training and query processes do not leak any private information. Moreover, based on the data aggregation matrix, multi-round interactions are not required in VFLR, improving training efficiency significantly. Detailed security analysis shows that VFLR can well protect data and model information from inference attacks. In addition, extensive experiments demonstrate that VFLR has high training and query accuracy and low computation and communication overhead.

The advance of cloud computing has driven an emerging trend of outsourcing the rapidly growing data and query services to a powerful cloud for easing the local storage and computing pressure. Meanwhile, when taking data privacy into account, data are usually outsourced to the cloud in an encrypted form. As a result, query services have to be performed over the encrypted data. Among all kinds of query services, the reverse kNN query is highly popular in various applications, such as taxi dispatching and targeted push of multimedia information, but its privacy has not received sufficient attention. To our best knowledge, many existing privacy-preserving reverse kNN query schemes still have some limitations on the query result accuracy, dataset privacy, and flexible support for the choice of the query object and the parameter k. Aiming at addressing these limitations, in this paper, we propose an efficient and privacy-preserving reverse kNN query scheme over encrypted data, named PRkNN. Specifically, we first design a modified M-tree (MM-tree) to index the dataset and further present an MM-Tree based reverse kNN query algorithm in the filter and refinement framework. Then, we leverage the lightweight matrix encryption to carefully design a filter predicate encryption scheme (FPE) and a refinement predicate encryption scheme (RPE); and propose our PRkNN scheme by applying them to protect the privacy of the MM-Tree based reverse kNN query algorithm. Detailed security analysis shows that FPE and RPE schemes are selectively secure, and our PRkNN scheme can preserve both query privacy and dataset privacy. In addition, we conduct extensive experiments to evaluate the performance of our scheme, and the results demonstrate that our scheme is efficient.

Location-based services (LBSs) provide enhanced functionality of mobile applications and convenience for mobile users, which plays a more and more remarkable role in people’s daily life. In LBSs, spatial range query is an essential tool for users to find interesting points in a specific region. However, during spatial range query, it is necessary for data owners and query users to exchange their location data, and the leakage of private location information has drawn significant attention in both governmental and social aspects. Meanwhile, most existing location privacy protection schemes only focus on achieving regular geometry range query over static location datasets. In this paper, we present an efficient and privacy-preserving arbitrary polygon range query scheme, named EPAPRQ. With EPAPRQ, the arbitrary and fine-grained polygon range query can be executed over a dynamic and time-series location dataset with privacy protection. Specifically, in EPAPRQ, an arbitrary polygon range query algorithm is first introduced with sub-range query technique. Then, to protect the private location information of the data owner and query users, a series privacy-preserving data computation protocols are constructed with a symmetric homomorphic encryption algorithm, and a ciphertext-based location dataset updating strategy is also designed. Finally, we propose a double filtration method through combining the quadtree index structure and minimum bounded rectangle, which is able to greatly improve the query efficiency over ciphertexts. Detailed security analysis shows that the sensitive location information in EPAPRQ can be well protected. Furthermore, we evaluate the performance of EPAPRQ in the real map, and the results demonstrate that EPAPRQ is indeed efficient.

Aiming at balancing data privacy and availability, Google introduces the concept of federated learning, which can construct global machine learning models over multiple participants while keeping their raw data localized. However, the exchanged parameters in traditional federated learning may still reveal the data information. Meanwhile, the training data are usually partitioned vertically in real-world scenes, which causes difficulties in model construction. To tackle these problems, in this paper, we propose an efficient and privacy-preserving vertical federated tree boosting framework, namely SGBoost, where multiple participants can collaboratively perform model training and query without staying online all the time. Specifically, we first design secure bucket sharing and best split finding algorithms, with which the global tree model can be constructed over vertically partitioned data; meanwhile, the privacy of training data can be well guaranteed. Then, we design an oblivious query algorithm to utilize the trained model without leaking any query data or results. Moreover, SGBoost does not require multi-round interactions between participants, significantly improving the system efficiency. Detailed security analysis shows that SGBoost can well guarantee the privacy of raw data, weights, buckets, and split information. Extensive experiments demonstrate that SGBoost can achieve high accuracy comparable to centralized training and efficient performance.

The advance of cloud computing has driven a new paradigm of outsourcing large-scale data and data-driven services to public clouds. Due to the increased awareness of privacy protection, many studies have focused on addressing security and privacy issues in outsourced query services. Although many privacy-preserving schemes have been proposed for various query types, the set reverse k nearest neighbors (RkNN) query is still an unexplored area. Even if some existing schemes can be adapted to achieve privacy-preserving set RkNN queries, they will suffer from linear search efficiency. As a steppingstone, in this paper, we propose an efficient and privacy-preserving set RkNN query scheme over encrypted data with sublinear query efficiency. Specifically, we first design an inverted prefix index to organize the set dataset and propose an algorithm to traverse the index with sublinear search efficiency. Then, we propose two oblivious data comparison protocols based on a symmetric homomorphic encryption (SHE) scheme and design the private filter/refinement protocols to preserve the privacy of index searching. After that, we propose an access pattern privacy-preserving set RkNN query scheme by using private filter/refinement protocols. Rigorous security analysis demonstrates that our scheme can protect data privacy and access pattern privacy. Experimental results indicate that our scheme is more efficient than the available naive solution in terms of computational costs and communication overheads.

The boom of cloud computing has stimulated the prevalence of outsourced query services, and privacy concerns further motivate extensive studies on privacy-preserving queries in the cloud. Graph similarity query is one critical query type, in which the similarity between two graphs is usually measured by graph edit distance (GED). Although many schemes have been proposed for GED computation/graph similarity query, they do not consider data privacy and are not applicable to the cloud computing scenario. To address this issue, in this paper, we propose the first efficient and privacy-preserving graph similarity query (PGSim) scheme in the filter and verification framework. Specifically, we first identify the pivot filter property of GED and use the property to design a pivot R-tree based filter algorithm, which can efficiently retrieve candidate graphs for graph similarity query. Then, we design a vertex mapping (VM) tree to index all vertex mappings between two graphs and develop a GED query verification algorithm to verify candidate graphs. After that, we design a suite of private algorithms based on a symmetric homomorphic encryption scheme and apply them to propose a pivot R-tree based filter predicate encryption (PRFilter) scheme and a private GED query verification (PGQVerify) algorithm. Based on the PRFilter scheme and the PGQVerify algorithm, we propose our PGSim scheme. Rigorous security analysis shows that our scheme is selectively secure. Performance evaluation also demonstrates the high efficiency of our scheme.

The aggregated reverse kNN (ARkNN) query aims to identify one query record with the maximum influence set and has become a powerful tool to support optimal decision-making in crowdsensing. Considering data privacy and query privacy, ARkNN queries should be performed in a private manner. Unfortunately, existing schemes cannot support privacy-preserving ARkNN queries over crowd-sensed data. To address this issue, we propose two efficient and privacy-preserving ARkNN query schemes with different security levels, named the BARQ scheme and the EARQ scheme, where the former can only protect data privacy while the latter can protect both data privacy and query privacy. Specifically, we first formalize the models of privacy-preserving ARkNN queries and propose our BARQ scheme based on a random response (RR) frequency oracle. Then, we design a privacy-preserving hardware-assisted reverse kNN query determination (PRkD) scheme for privately determining whether a query record is among the RkNN of a data record. After that, we present our EARQ scheme by leveraging the PRkD scheme to protect query privacy and integrating the RR frequency oracle to protect data privacy. In addition, our rigorous security analysis demonstrates that the BARQ scheme can well protect data privacy, and the EARQ scheme can protect both data privacy and query privacy. Extensive experimental results illustrate that they have high accuracy in query results and are efficient in computational costs and communication overheads.

Reverse k nearest neighbor (RkNN) query has been widely applied in the targeted push of information. Many schemes for the RkNN query on encrypted data have been proposed for coordinating the emerging trend of outsourcing data to the cloud. However, none of them supports the spatial data with many features, a prevalent data type in location-based services, e.g., each user in online dating apps usually has a spatial location and many personality trait features. Meanwhile, incorporating features with the spatial data endows the spatial-feature-based RkNN query to provide more precise services than the spatial-based RkNN query. Therefore, as a steppingstone, we propose an efficient and privacy-preserving spatial-feature-based RkNN scheme in this work for the first time. Specifically, we first design a modified intersection and union R tree (MIUR-tree) to index the spatial and feature data. Then, we introduce an MIUR-tree based RkNN query algorithm in the filter and refinement framework to efficiently process RkNN queries. After that, based on a symmetric homomorphic encryption (SHE) scheme, we design a private filter protocol and a private refinement protocol, and leverage them to propose our RkNN query scheme. Rigorous security analysis demonstrates that our scheme is privacy-preserving, and extensive experiments indicate that our scheme is computationally efficient.

The advance of artificial intelligence has stimulated the rapid development of multi-biometric recognition that recognizes users based on their multiple biometric features. Since biometric features are private information of users, multi-biometric recognition has to be processed in a privacy-preserving way. Although many approaches have been proposed for privacy-preserving multi-biometric recognition, they have some limitations in security and performance. Aiming at this challenge, we propose a privacy-preserving multi-biometric recognition scheme. In this scheme, we first utilize a neural network to train a face and voiceprint fusion model, which obtains better fusion layer parameters and improves the accuracy of fusion recognition. Then, we design an efficient and privacy-preserving multi-biometric recognition scheme based on the MK-CKKS cryptosystem in the decentralized model, which does not rely on a trusted third party to increase its universality. The scheme can guarante recognition accuracy by utilizing a feature-level fusion mechanism, reaching EERs as low as 0.66%. Security analysis shows that the privacy of biometric information and recognition results can be guaranteed. Experimental results validates the efficiency of our scheme.

Weighted similarity query has been an essential primitive to enable personalized disease diagnosis in eHealthcare. With the prevalence of cloud computing, a new paradigm is to outsource weighted similarity range query services to the cloud. Meanwhile, the query services are usually processed over encrypted data considering data privacy. Although many existing schemes are available to achieve privacy-preserving weighted similarity query over encrypted data, they have some security and query efficiency drawbacks. This paper addresses this problem by proposing an efficient and privacy-preserving weighted similarity range query scheme. First, we employ a k-d tree to index the outsourced dataset and present a k-d tree based weighted similarity range query algorithm. Then, we propose our scheme by applying the MASPE scheme to protect the privacy of the k-d tree based weighted similarity queries. Privacy preservation of our scheme is proved through security analysis. Efficiency improvement is confirmed by the extensive experiments that indicate that our scheme improves 7 x query efficiency than the state-of-the-art scheme.

As a straightforward consequence of advances in the Internet of Things (IoT), location-based service (LBS) applications have been pervasive in our daily lives. Nevertheless, since those LBS applications will continuously collect and disclose users’ location data, major concerns on privacy leakage are raised. Aiming at the challenge, in this article, we first build up a detect module (DM) and employ it to investigate more than 80% of LBS applications are keen on tracking users. Then, to thwart the threats from those LBS applications, we exploit the deceptive dummy techniques and design a dummy-based location privacy preserving scheme, named dummy location provider (DLP), which comprises three algorithms, namely, Spread, Shift, and Switch. Specifically, Spread and Shift are in charge of generating deceptive dummies and trajectories. And with Switch, users’ real locations are replaced with dummy trajectories before being submitted to LBS applications. As a result, users can not only prevent applications from accessing location data arbitrarily, but also avoid being questioned by applications in terms of honesty. Furthermore, to guarantee necessary functions of LBS, DLP offers customizable privacy-preserving strategies for users, which can achieve flexible location data usage control. Finally, our DLP can also attain achievable and effortless deployment over smart devices. Detailed security analysis indicates that DLP resists inference attacks even facing skeptical applications. In addition, for performance evaluation, a DLP application (DLPA) is developed on the Android platform and tested in the real environment, and the extensive experimental results demonstrate that the DLPA is indeed effective and high efficiency in practice.

Mobile crowdsensing (MCS) is a newly emerged sensing paradigm, where a large group of mobile workers collectively sense and share data for real-time services. However, one major problem that hinders the further development of MCS is the potential leakage of workers’ data privacy. In this article, we integrate federated learning (FL) with MCS and introduce a novel sensing system, called federated MCS (F-MCS). In F-MCS, the workers can optimize the global model while keeping all the sensitive training data locally, thus ensuring their data privacy. Nevertheless, there are still two major issues in F-MCS. The first issue is that in F-MCS services, the workers are heterogeneous in terms of computational capacities and data resources. Hence, qualified workers should be appropriately selected to improve the efficiency of the training process. The second issue is that F-MCS is across-deviceFL system, where the platform will finally get the global model after multiple training rounds. However, most privacy-preserving techniques are designed forcross-siloFL platforms, which cannot be applied to real-world F-MCS scenarios. To tackle the above problems, in this article, we propose a privacy-preserving scheme for F-MCS, namely, FedSky. Mainly, by extending the classic FedAvg algorithm, FedSky selects qualified workers based on the constrained group skyline (CG-skyline) and securely aggregates model updates based on the homomorphic encryption technique. Comprehensive security analysis demonstrates the privacy preservation of FedSky. Extensive experiments are conducted on an image classification task, where the comparison results validate the proposed scheme’s efficiency and effectiveness.

Healthcare data explosion and cloud computing booming have motivated healthcare centers to outsource their healthcare data and data-driven services to a powerful cloud. Nevertheless, due to privacy concerns, the data are usually encrypted before being outsourced, which will degrade the data utility and make it challenging to implement data-driven services. Although the multidimensional range query over encrypted data, as one of the most popular outsourced services in eHealthcare, has been extensively studied, existing solutions still have some limitations in efficiency, privacy, and practicality. Aiming at this challenge, in this article, we design an efficient and privacy-preserving multidimensional range query (PMRQ) scheme. We first build an R-tree to index the data set and reduce the R-tree-based range queries to the multidimensional range intersection problem. Then, by delicately designing a data comparison algorithm and a homomorphic encoding technique, we present an encoding-based range intersection algorithm. After that, by employing matrix encryption to protect the privacy of the encoding-based range intersection algorithm, we design a multidimensional range intersection predicate encryption (MRIPE) scheme. Based on the MRIPE scheme, we then propose our PMRQ scheme. A detailed security analysis illustrates that our PMRQ scheme is privacy preserving, and experimental results demonstrate that it is computationally efficient.

It is believed that online disease risk assessment system has great potential to alleviate the medical treatment problems for the future smart city and communities, as it can excavate disease risk factors from a large number of patient features, provide diagnostic references for doctors, and save medical treatment time for patients. However, the flourish of online disease risk assessment service still faces severe challenges including information privacy and security. In this article, based on the naïve Bayesian classification, we propose an efficient and privacy-preserving diseaserisk assessment scheme over multi-outsourced vertical datasets, named CARER. With CARER, the e-healthcare provider can securely train a disease risk predication model over vertically distributed medical data from multiple medical centers (i.e., hospitals), and provide privacy-preserving disease risk predication services for users (i.e., patients and doctors). During the model training and disease risk prediction phases, all sensitive data are operated over ciphertexts without decryption. As a result, the private information of medical centers, e-healthcare provider, and users can be well protected. Detailed security analysis shows that CARER can resist various known security threats. In addition, we evaluate the performance of CARER with real medical datasets, and the results demonstrate that CARER is efficient.

Over the past years, the increasingly severe data island problem has spawned an emerging distributed deep learning framework—federated learning, in which the global model can be constructed over multiple participants without directly sharing their raw data. Despite its promising prospect, there are still many security challenges in federated learning, such as privacy preservation and integrity verification. Furthermore, federated learning is usually performed with the assistance of a center, which is prone to cause trust worries and communicational bottlenecks. To tackle these challenges, in this paper, we propose a privacy-preserving and verifiable decentralized federated learning framework, named PVD-FL, which can achieve secure deep learning model training under a decentralized architecture. Specifically, we first design an efficient and verifiable cipher-based matrix multiplication (EVCM) algorithm to execute the most basic calculation in deep learning. Then, by employing EVCM, we design a suite of decentralized algorithms to construct the PVD-FL framework, which ensures the confidentiality of both global model and local update and the verification of every training step. Detailed security analysis shows that PVD-FL can well protect privacy against various inference attacks and guarantee training integrity. In addition, the extensive experiments on real-world datasets also demonstrate that PVD-FL can achieve lossless accuracy and practical performance.

Speaker verification has played a significant role in authentication with the booming development of smartphones and intelligent terminals in recent years. However, most speaker verification systems directly store the users original voiceprint template data (or called acoustic features). In this paper, we reveal the insecurity and sensitiveness of voiceprint template data by carrying out spoofing attacks on speaker verification systems using genetic algorithm. Meanwhile, multiple generation models based on different genetic algorithms (standard genetic algorithm, multiple population genetic algorithm) are proposed, but also the effects of these generation models are compared. Moreover, experimental results on state-of-the-art text-independent speaker verification techniques (such as i-vector, GMM-UBM) clearly demonstrate that our generated attack voice with leaked voiceprint template data can completely imitate users and pass the speaker verification.

Online medical primary diagnosis system, which can provide convenient medical decision support through applying mobile communication and data analysis technology, has been considered as a promising approach to improve the quality of healthcare service. However, it still faces many severe challenges on the privacy of users' health information and the accuracy of diagnosis result, which deter the wide adoption of online medical primary diagnosis system. In this paper, we propose an efficient and privacy-preserving online medical primary diagnosis (CINEMA) framework. Within CINEMA framework, users can access online medical primary diagnosing service accurately without divulging their medical data. Specifically, based on fast secure permutation and comparison technique, the encrypted user's query is directly operated at the service provider (SP) without decryption, and the diagnosis result can only be decrypted by the user, meanwhile, the diagnosis model in SP can also be protected. Through extensive analysis, we show that CINEMA can ensure that user's health information and healthcare SP's diagnosis model are kept confidential, and has significantly reduce computation and communication overhead. In addition, performance evaluations via implementing CINEMA demonstrate its effectiveness in term of the real environment.

Online medical primary diagnosis system, which can provide the pre-diagnosis service anywhere anytime, has attracted considerable interest. However, the flourish of online medical primary diagnosis system still faces many serious challenges since the sensitivity of personal health information and service provider''s diagnosis model. In this paper, we propose an efficient and privacy-preserving medical primary diagnosis scheme based on k-nearest-neighbors classification (kNN), called EPDK. With EPDK, medical users can ensure that their sensitive health information are not compromised during the online medical diagnosis process, and service provider can provide high-accuracy service without revealing its diagnosis model. Specifically, based on lightweight multiparty random masking and polynomial aggregation techniques, a medical user preprocesses her/his query vector before sending out and the preprocessed vector is directly operated in the service provider without obtaining original data, meanwhile, the primary diagnosis result cannot be achieved by anyone except the medical user. Through extensive analysis, we show that EPDK can resist multifarious known security threats, and has significantly lower computation complexity than existing schemes. Moreover, performance evaluations via implementing EPDK in the real environment demonstrate that EPDK is highly efficient in terms of computation overhead.

With the pervasiveness of location-aware mobile terminals and the popularity of social applications, location-based social networking service (LBSNS) has brought great convenience to people's life. Meanwhile, proximity detection, which makes LBSNS more flexible, has aroused widespread concern. However, the prosperity of LBSNS still faces many severe challenges on account of users' location privacy and data security. In this paper, we propose two efficient and privacy-preserving proximity detection schemes, named arbitrary geometric range query for polygons (AGRQ-P) and arbitrary geometric range query for circles (AGRQ-C), for location-based social applications. With proposed schemes, a user can choose any area on the map, and query whether her/his friends are within the region without divulging the query information to both social application servers and other users, meanwhile, the accurate locations of her/his friends are also confidential for the servers and the query user. Specifically, with algorithms based on ciphertext of geometric range query, users' query and location information is blurred into ciphertext in client, thus no one but the user knows her/his own sensitive information. Detailed security analysis shows that various security threats can be defended. In addition, the proposed schemes are implemented in an IM APP with a real LBS dataset, and extensive simulation results over smart phones further demonstrate that AGRQ-P and AGRQ-C are highly efficient and can be implemented effectively.