Privacy-preserving federated learning (PPFL) is a strong-privacy distributed learning paradigm, which typically employs secure aggregation (SA) protocols to protect the aggregation results of federated learning. However, the high computational cost of existing SA protocols is difficult to generalize to decentralized federated learning (DFL) with the large number of clients and fails to defend against model poisoning attacks launched by Byzantine adversaries. In this paper, we propose an efficient computational SA protocol compatible with DFL, referred to as PriDFL, and address the issue of Byzantine-robust aggregation. Specifically, we design an advanced secret-sharing protocol based on number-theoretic transformations to reduce the computational complexity from$O(n^{2})$to$\mathcal {O}(n\log n)$during data sharing. We employ a single-mask approach to provide lightweight gradient privacy protection for DFL. To mitigate the impact of poisoned gradients on model convergence, we develop a Byzantine resilience criterion grounded in model cross-updating. The proposed criterion efficiently detects poisoned gradients and non-independent identically distributed (non-IID) data with local computation. Security analysis shows that PriDFL satisfies the security requirements in an honest but curious setting. We evaluate PriDFL on typical datasets (e.g., MNIST and CIFAR-10) and the results show that PriDFL is computation-communication efficient and Byzantine resilient. In particular, PriDFL optimizes the computational efficiency by 6-10× compared to the well-utilized SA protocols while supporting Byzantine robustness.

Connected autonomous vehicles (CAVs) utilize multi-modal sensors, such as LiDAR and high-definition cameras, to collect diverse types of sensing data. Fusing object detection information from these two modalities facilitates more accurate environmental perception. In this context, lightweight secret sharing techniques are employed to protect information privacy, enabling further calculation while effectively alleviating the computational resource constraints of CAVs. Meanwhile, such techniques require an additional third-party to generate some necessary random numbers. Addressing the challenges of privacy disclosure of multi-modal object information and the reliability of random numbers, we propose a malicious third-party-resistant privacy-preserving multi-modal object fusion model, termed MPOF. First, we develop a series of secure computation protocols that do not rely on time-consuming cryptographic primitives, including secure multiplication, secure sharing conversion, and secure comparison. Leveraging the idea of sacrificial verification, we can effectively detect malicious behavior by the third-party during the random number generation process. Second, we construct a secure object bounding-box matching module based on arithmetic secret sharing (ASS), enabling similarity calculation and matching of bounding-boxes between point cloud and image modalities. Additionally, we design a secure object score fusion module that achieves fusion and updating through secure implementations of convolution, ReLU, and Maxout operations. Detailed theoretical analysis and experimental results demonstrate that, compared to secure computation protocols using homomorphic encryption for random number generation, the proposed protocols reduce computational overhead by five orders of magnitude. Furthermore, the MPOF model constructed by integrating these protocols is secure, accurate, and efficient.

In mobile crowdsensing (MCS) applications, the single type data is inadequate to reflect the complexities of the real world and meet precise task requirements. Currently, there are few works that focus on mixed data in the context of MCS, and there is no work considering the credit issues of sensing platform. The privacy, fairness, and reliability of assessing the quality of mixed data remain unguaranteed. Therefore, we design a high-efficiency and privacy-preserving mixed data quality assessment scheme which adopts a dual-server architecture, designs secure k-prototype clustering for quality assessment, and conducts anomaly detection to eliminate anomalous data. Furthermore, we design a fair and reliable allocation mechanism to fairly allocate reward to users based on fixed and floating reward mechanisms for incentivizing rational users to submit high-quality mixed data. To prevent payment defaults by the sensing platform, we design verifiable credential to restrict them, ensuring payment fairness and transactional reliability. Finally, through theoretical analysis and experimental evaluation, we demonstrate the effectiveness and security of the proposed scheme. The results indicate that in terms of efficiency, the time overhead of mixed data quality assessment has been significantly reduced by three orders of magnitude compared to existing schemes.

Secure inference in a deep-learning-as-a-service setting (DLaaS) can effectively protect sensitive data of the client and server model parameters. However, various nonlinear computations heavily hinder its efficiency. To address this issue, we propose a secure three-party inference framework, called LSTN, to ensure the privacy of client input data and meanwhile achieve prediction accuracy close to the plaintext setting. Specifically, we leverage replicated secret sharing to design a novel secure three-party comparison protocol that will be employed to develop a secure ReLU function. Our developed protocol can achieve high communication efficiency in the scenario of having a majority of honest parties. The experimental result shows that the inference time is 6× faster than the prevailing computing framework, CrypTen.

Applying knowledge distillation (KD) in federated learning (FL) can transfer model knowledge between clients’ local models and global model, which helps to improve the generalization of the global model. However, this requires both the clients and the server to have public data sets, which may lead to potential privacy disclosure issues. In this paper, we propose a federated data-free knowledge distillation framework, namely FedDFKD, which does not rely on any public data sets. There is a lightweight delivery model we design to learn and transfer model knowledge in different clients. During local training, the local model is jointly trained with delivery model using local data sets, and the local model feeds back its knowledge to the delivery model after it has finished its training phase in this communication round. Afterwards, the server performs global model aggregation and knowledge distillation of the delivery model. Finally, the server returns global model and distillation result to clients. We compare FedDFKD with the most representative aggregation algorithms in FL, and the results show that our method is feasible and outperforms the compared methods by between 0.1 and 3.96 percent of the global model on the MNIST dataset.

Federated learning of deep neural networks has emerged as an evolving paradigm for distributed machine learning, gaining widespread attention due to its ability to update parameters without collecting raw data from users, especially in digital healthcare applications. However, the traditional centralized architecture of federated learning suffers from several problems (e.g., single point of failure, communication bottlenecks, etc.), especially malicious servers inferring gradients and causing gradient leakage. To tackle the above issues, we propose a robust and privacy-preserving decentralized deep federated learning (RPDFL) training scheme. Specifically, we design a novel ring FL structure and a Ring-Allreduce-based data sharing scheme to improve the communication efficiency in RPDFL training. Furthermore, we improve the process of distributing parameters of the Chinese residual theorem to update the execution process of the threshold secret sharing, supporting healthcare edge to drop out during the training process without causing data leakage, and ensuring the robustness of the RPDFL training under the Ring-Allreduce-based data sharing scheme. Security analysis indicates that RPDFL is provable secure. Experiment results show that RPDFL is significantly superior to standard FL methods in terms of model accuracy and convergence, and is suitable for digital healthcare applications.

Well-trained neural network models are deployed on edge servers to provide valuable inference services for clients. To protect data privacy, a promising way is to exploit various types of secret sharing to implement privacy-preserving neural network inference. However, existing schemes suffer high communication rounds and overhead, making them hardly practical. In this paper, we propose Cenia, a new communication-efficient privacy-preserving neural network inference model. Specifically, we exploit arithmetic secret sharing to develop low-interaction secure comparison protocols, that can be used to realize secure activation layers (e.g., ReLU) and secure pooling layers (e.g., max pooling) without expensive garbled circuit and oblivious transfer primitives. Besides, we also design secure exponent and division protocols to realize secure normalization layers (e.g., Sigmoid). Theoretical analysis demonstrates the security and low complexity of Cenia. Extensive experiments have also been conducted on benchmark datasets and classical models, and experimental results show that Cenia achieves privacy-preserving, accurate, and efficient neural network inference. Particularly, Cenia can achieve 37.5% and 60.76% of Sonic’s communication rounds and overhead, respectively, compared to Sonic (i.e., the state-of-the-art scheme).

Outsourced cloud computing can be considered as an effective way to overcome the data island among users and relieve the pressure of limited resources. However, due to the concerns about trust in cloud servers, outsourcing the users’ data and model training task has considerable privacy disclosure risks. This article presents a PriKPM scheme by using additive secret sharing (ASS), so as to implement the privacy-preserving${k}$-prototype clustering for mixed data (i.e., including numerical and categorical attributes). In PriKPM, data samples are randomly split into two shares and delivered offline to two collaborative servers. We design a secure initialization method for determining the location and number of cluster centers. Then, both servers securely calculate the mixed distance between samples and cluster centers, and execute the samples partion and cluster updating operations. An efficient and secure comparison protocol is developed to offer flexibly the “less than or equal” and “equal” functions during the entire clustering process. Furthermore, theoretical analysis proves the effectiveness and security of PriKPM. Sufficient experiments demonstrate that PriKPM is computationally more efficient than existing secure clustering works. PriKPM can achieve the approximate accuracy of the plaintext${k}$-prototype clustering scheme.

Connected autonomous vehicles (CAVs) are capable of capturing high-definition images from onboard sensors, which can be used to facilitate the detection of objects in the vicinity. Such images may, however, contain sensitive information (e.g., human faces and license plates) as well as the indirect location of CAVs. To protect the object privacy of images shared by CAVs, this article proposes a privacy-preserving object detection (P2OD) framework. Specifically, we propose multiple secure computing protocols designed to construct a privacy-preserving Faster$R$-convolutional neural network (CNN) model to securely extract features and bounding-boxes of objects in an image. By leveraging edge computing (with higher performance computation and lower latency, in comparison to cloud-based solutions), CAVs randomly split the captured images and upload them to two noncollusive edge servers. Both servers will then perform the P2OD framework cooperatively to directly detect objects over random image shares without exposing sensitive information. The theoretical analysis demonstrates the security, correctness, and efficiency of the P2OD framework, and the experimental findings show that the P2OD framework can effectively protect the classification and location privacy of image objects for CAVs. Compared with the original Faster R-CNN model, the classification and regression errors of the P2OD framework can be controlled within 10−12 and 10−14, respectively.

Aiming to the data and model privacy issue in outsourced clustering tasks, this paper proposes an practical privacy-preserving k-prototype clustering scheme (referred to PriKPM) supporting mixed numerical and categorical attributes data. In PriKPM scheme, the users only randomly split the data sample into two shares and send them to two non-collusive servers, without interacting with the servers online. The two servers can cooperate to perform secure sample distance calculation, cluster center selection, and in-cluster sample update operations over two randomness data shares, and finally obtain the clustering distribution of samples. Specifically, we design an efficient secure comparison protocol based on additive/arithmetic secret sharing, which can switch freely between "greater than or equal" and "equal" functions, providing two comparison forms for PriKPM scheme. Theoretical analysis indicates the security and efficiency of our PriKPM scheme. Experimental results further show that compared to prior work, the clustering time of PriKPM scheme is reduced by 3 orders of magnitude.

Collaborative perception enables autonomous vehicles to exchange sensor data among each other to achieve cooperative object classification, which is considered an effective means to improve the perception accuracy of connected autonomous vehicles (CAVs). To protect information privacy in cooperative perception, we propose a lightweight, privacy-preserving cooperative object classification framework that allows CAVs to exchange raw sensor data (e.g., images captured by HD camera), without leaking private information. Leveraging chaotic encryption and additive secret sharing technique, image data are first encrypted into two ciphertexts and processed, in the encrypted format, by two separate edge servers. The use of chaotic mapping can avoid information leakage during data uploading. The encrypted images are then processed by the proposed privacy-preserving convolutional neural network (P-CNN) model embedded in the designed secure computing protocols. Finally, the processed results are combined/decrypted on the receiving vehicles to realize cooperative object classification. We formally prove the correctness and security of the proposed framework and carry out intensive experiments to evaluate its performance. The experimental results indicate that P-CNN offers exactly almost the same object classification results as the original CNN model, while offering great privacy protection of shared data and lightweight execution efficiency.

Connected autonomous vehicles (CAVs) employ the point cloud data captured by LiDAR to enhance the capability of object recognition and detection. Edge computing with its inherent advantages can help CAVs alleviate resource constraints and enable faster situational awareness and data processing. However, the point cloud data contains private information, such as vehicle identity, location and trajectory, directly uploading the raw point cloud to the edge nodes or other vehicle will lead to serious privacy leakage. To the best of our knowledge, we are the first to try to tackle this challenge and propose a privacy-preserving object detection framework over random point cloud shares for CAVs (referred to SecPCV), aiming to guarantee the privacy of both point cloud and object detection results. In SecPCV, CAVs split point cloud into two random shares based on additive secret sharing (ASS) and upload them to two competing edge nodes, respectively, which greatly compress the computational load of CAVs. Without changing the object detection network in plaintext environment, the edge nodes can cooperatively and securely extract, regress, and classify over point cloud shares. Theoretical analysis ensure the efficiency and security of the SecPCV framework. Experimental results with the real KITTI point cloud dataset indicate that SecPCV can achieve the consistent object detection accuracy as that in plaintext environment, and provide a feasible solution for CAVs secure sharing of point cloud data.

It is enormously challenging to achieve a satisfactory balance between quality of service (QoS) and users’ privacy protection along with measuring privacy disclosure in social Internet of Things (IoT). We propose a privacy-preserving personalized service framework (Persian) based on static Bayesian game to provide privacy protection according to users’ individual security requirements in social IoT. Our approach quantifies users’ individual privacy preferences and uses fuzzy uncertainty reasoning to classify users. These classification results facilitate trustworthy cloud service providers (CSPs) in providing users with corresponding levels of services. Furthermore, the CSP makes a strategic choice with the goal of maximizing reputation through playing a decision-making game with potential adversaries. Our approach uses Shannon information entropy to measure the degree of privacy disclosure according to the probability of game mixed strategy equilibrium. Experimental results show that Persian guarantees QoS and effectively protects user privacy despite the existence of adversaries.