Privacy-preserving Transformer inference has gained attention due to the potential leakage of private information. Despite recent progress, existing frameworks still fall short of practical model scales, with gaps up to a hundredfold. A possible way to close this gap is the Mixture of Experts (MoE) architecture, which has emerged as a promising technique to scale up model capacity with minimal overhead. However, given that the current secure two-party (2-PC) protocols allow the server to homomorphically compute the FFN layer with its plaintext model weight, under the MoE setting, this could reveal which expert is activated to the server, exposing token-level privacy about the client's input. While naively evaluating all the experts before selection could protect privacy, it nullifies MoE sparsity and incurs the heavy computational overhead that sparse MoE seeks to avoid. To address the privacy and efficiency limitations above, we propose a 2-PC privacy-preserving inference framework, SecMoE. Unifying per-entry circuits in both the MoE layer and piecewise polynomial functions, SecMoE obliviously selects the extracted parameters from circuits and only computes one encrypted entry, which we refer to as Select-Then-Compute. This makes the model for private inference scale to 63× larger while only having a 15.2× increase in end-to-end runtime. Extensive experiments show that, under 5 expert settings, SecMoE lowers the end-to-end private inference communication by 1.8~7.1× and achieves 1.3~3.8× speedup compared to the state-of-the-art (SOTA) protocols.

Linkable Ring Signatures (LRS) allow users to anonymously sign messages on behalf of ad-hoc rings, while ensuring that multiple signatures from the same user can be linked. This feature makes LRS widely used in privacy-preserving applications like e-voting and e-cash. To scale to systems with large user groups, efficient schemes with short signatures and fast verification are essential. Recent works, such as DualDory (ESORICS'22) and LLRing (ESORICS'24), improve verification efficiency through offline precomputations but rely on static rings, limiting their applicability in ad-hoc ring scenarios. Similarly, constant-size ring signature schemes based on accumulators face the same limitation.

Fully homomorphic encryption (FHE) enables computation directly over encrypted data without decryption, offering a promising approach to privacy-preserving outsourcing computation in cloud environment. However, its high computational cost, especially for some fundamental operations such as matrix multiplication, remains a major obstacle to practical deployment. Although several homomorphic matrix multiplication schemes have been proposed for acceleration, they still suffer from excessive and costly multiplication and rotation operations. In this work, we propose an efficient and generic homomorphic matrix multiplication scheme using the matrix complexification technique and the Baby-Step Giant-Step (BSGS) strategy. Matrix complexification halving the number of multiplications by mapping matrix elements into complex numbers, allowing each homomorphic complex multiplication to process two products simultaneously. Separately, BSGS halving the number of rotation in alignment phase through a hierarchical rotation scheme involving fine-grained pre-rotations followed by coarse-grained main rotations. We implement our approach using the HEaaN library and evaluate it across a range of matrix dimensions. Results show a 34%–68% runtime reduction over prior state-of-the-art methods. Our method advances the efficiency and scalability of encrypted matrix computation, making FHE more viable for real-world applications.

The reliability and robustness of lane detection play an instrumental role in the practical deployment of autonomous driving systems. Despite previous research indicating that adversarial examples can negatively affect lane detection models, leading to erroneous lane detection, most existing adversarial attacks are designed to attack lane detection models of a single deep learning-based task. To address this issue, we propose a towards cross-task physical adversarial attack via LED illumination modulation for lane detection systems (AdvLIM-LD) by analyzing the common critical feature response mechanisms of lane detection models of different task types. The method first introduces a CFD module to evaluate the impact of adversarial examples on critical features contributing to various tasks. Then, it utilizes the pulse width modulation of LEDs and the rolling shutter effect of CMOS image sensor to implant imperceptible adversarial perturbations during the image acquisition process. Finally, it enhances the cross-task transferability of adversarial examples by jointly optimizing a dual objective function composed of critical feature loss and model output accuracy loss. In digital domain simulations, adversarial examples generated by AdvLIM-LD caused an average detection accuracy degradation of 74.64% across lane detection models representing different task types, including the segmentation model SCNN, the anchor-based detector LaneATT, the curve-fitting model LSTR, and the keypoint detection model GANet. For cross-task attacks, the average accuracy degradation reached 55.37%. Physical-world experiments further validated the effectiveness of AdvLIM-LD, demonstrating its ability to degrade the source models’ average detection accuracy by over 80%.

Blockchain-based distributed mobile crowdsensing (MCS) has been widely adopted in areas, such as the Industrial Internet of Things (IIoT) to transform traditional data collection methods. However, existing works lack a theory-driven quantitative utility analysis and precise reward and punishment measures that effectively deter malicious behavior. This deficiency hinders the ability to constrain rational participants’ behavior, thereby compromising the trustworthiness of MCS implementations. To address the issue above, this article proposes Crowd-BT: a bilateral trustworthy ensured scheme for blockchain assisted MCS. Specifically, Crowd-BT first devises reward and punishment measures based on participants’ short-term utility analysis for an unrestricted context in the ideal case. Then, for the more practical situation, Crowd-BT designs blocklist punishment measures for participants, ensuring self-containment based on long-term utility analysis. Finally, leveraging the practical blocklist punishment measures and the devised anonymity behavior record certificates (ABRCs) using the message-hidden signature, Crowd-BT details the implementation steps for MCS under the blockchain. Theoretical analysis shows that Crowd-BT achieves both utility and security goals. Extensive experiments illustrate that Crowd-BT effectively constrains participant behavior. Compared to the existing works, it significantly enhances the trustworthiness of MCS.

Although face recognition starts to play an important role in our daily life, we need to pay attention that data-driven face recognition vision systems are vulnerable to adversarial attacks. However, current digital adversarial attacks and physical adversarial attacks both have drawbacks, with the former ones impractical and the latter one conspicuous, high-computational and low-executable. To address the issues, we propose a practical, executable, stealthy and low computational adversarial attack based on LED illumination modulation. To fool the systems, the proposed attack generates physically imperceptible luminance changes to human eyes through fast intensity modulation of scene LED illumination and uses the rolling shutter effect of CMOS image sensors in face recognition systems to implant luminance information perturbation to the captured face images. In summary, we present a denial-of-service (DoS) attack for face detection and an evasion attack for face verification. We also evaluate their effectiveness against wellknown face detection models, Dlib, MTCNN and RetinaFace, and face verification models, Dlib, FaceNet, and ArcFace. The extensive physical experiments show that the success rates of DoS attacks against face detection models reach 97.67%, 100%, and 100%, respectively, and the success rates of evasion attacks against all face verification models reach 100%.

General matrix multiplication (GEMM) as a fundamental subroutine has been widely used in many applications like scientific computing, machine learning, etc. Although many studies are dedicated to optimizing its performance, they mainly focus on matrices with regular shapes or x86 platforms. The irregularly shaped matrices on GEMM running on modern ARMv8 processors are under-explored. In this paper, we provide a thorough performance analysis of the general block-panel multiplication (GEBP) kernel of GEMM that has irregular shapes. Based on our analysis, we propose a new GEMM algorithm named EPPA with three novel schemes to improve GEMM performance on ARMv8 processors: i) eliminating packing to reduce Ll cache contention, ii) avoiding data eviction and pre-fetching data to reduce the Ll cache miss penalty, and iii) an adaptive selection strategy of the above two and original schemes. We conduct extensive experiments with a large range of irregular matrices on three popular ARMv8 processors compared to seven state-of-the-art GEMM libraries. The experimental results show that our EPPA algorithm outperforms existing ones across workloads and processors and accelerates real-world applications.

The rapid proliferation of Internet of Things (IoT) devices has led to an increased need for robust and efficient intrusion detection systems capable of identifying and mitigating novel threats. Traditional methods often struggle with the scarcity of labeled anomaly data, which is highly consequential, particularly in the context of IoT. In this study, we propose a novel few-shot learning approach by leveraging a Multi-Stage Attention Siamese Network (MASiNet) for network traffic intrusion detection based on meta-learning framework. Unlike traditional methods, the proposed MASiNet model is capable of detecting intrusions with minimal labeled samples, addressing the challenge of scarce anomaly data. The model is trained using various attack samples and evaluates unknown samples by comparing similarities with a small set of known attack types. A well-structured cost function design, incorporating two specific losses, is introduced to optimize the effectiveness of the training process. Tested on the NSL_KDD and UNSW-NB15 datasets in a simulated few-shot learning environment, the MASiNet model demonstrates superior performance in terms of accuracy, precision, False Alarm Rate (FAR), outperforming existing methods. Furthermore, we have validated our approach through real-world evaluations. The proposed method provides an effective solution for intrusion detection in the context of few-shot learning, offering a proficient solution that aligns with the dynamic nature of IoT networks.

Compared to conventional wireless body area networks (WBANs), the amount of data processed and the analytical capabilities offered by cloud-based WBANs are significantly more extensive. Nevertheless, the paramount consideration in such contexts remains the security and privacy ramifications. Concurrently, the process where medical cloud server (MCS) computes the response aggregation data may be opaque and there is a risk that (partially) invalid aggregation results may be presented to the task requester, either intentionally (e.g., malicious or cost-saving) or unintentionally (e.g., corruption or processing error). Furthermore, with the different roles played by each data requester, relying solely on a single data aggregation type is no longer sufficient to satisfy the diverse data aggregation requests from these requesters. To this end, this paper proposes a novel verifiable multi-dimensional encrypted medical data aggregation scheme (VMEMDA) for cloud-based WBANs, where we integrate an extended super-increasing sequence with a modified Paillier cryptosystem. Doing so allows us to ensure that each dimensional medical data collected by wireless sensor devices and corresponding square values can be encrypted into a single ciphertext with the chronological time series. This enables MCS to select various aggregation types, such as spatial/temporal data aggregation, to aggregate the multi-source encrypted medical data into a single ciphertext. Then the task requester can conduct diverse privacy-preserving statistical analyses, including sum, average, and variance. Moreover, we utilize a homomorphic hash function to guarantee the encrypted data integrity in a highly efficient way, and we design an unpredictable random sequence and integrate it into the provable data possession mechanism to achieve aggregated data correctness guarantee. Performance evaluation demonstrates that VMEMDA exhibits considerably lower computation and communication overhead compared to other existing multi-dimensional data aggregation schemes.

The fully homomorphic encryption schemes FHEW/TFHE support fast bootstrapping to refresh ciphertexts. However, executing FHEW/TFHE with SIMD bootstrapping is inefficient due to the algebraic structure. For the theoretical state of the art, LW23 (Liu and Wang, Eurocrypt2023) proposed a mathematical framework for SIMD bootstrapping, while the ciphertext slots are limited in size $\tilde O\left( {{\lambda ^{0.25}}} \right)$ and total complexity of $O\left( {{2^{\rho - 1}}n} \right)$ for n(= O(λ)) LWE ciphertexts.This paper aims at developing an efficient FHE scheme with SIMD bootstrapping, called Batch RGSW, which yields overall bootstrapping complexity O(n2/ε), ε > 2. To achieve this, there are two main steps to craft our scheme. Firstly, we decompose a cyclotomic ring into two smaller subrings R1 for the ciphertext space and ℛ2for the base as slots. The ciphertext is then extracted with basic automorphic rotations. In this way, we obtain a batch RGSW scheme with ciphertext slots in size of $\tilde O\left( {{\lambda ^{0.5}}} \right)$. Secondly, we combine Partial Fast Fourier Transformation (PFFT) with Batch RGSW scheme to reduce bootstrapping amortized cost. We utilize PFFT for coefficient decomposition and perform batch homomorphic multiplications. We end up with a bootstrapping algorithm with amortized complexity O(1).

Abstract Federated learning (FL), a variant of distributed learning (DL), supports the training of a shared model without accessing private data from different sources. Despite its benefits with regard to privacy preservation, FL’s distributed nature and privacy constraints make it vulnerable to data poisoning attacks. Existing defenses, primarily designed for DL, are typically not well adapted to FL. In this paper, we study such attacks and defenses. In doing so, we start from the perspective of DL and then give consideration to a real-world FL scenario, with the aim being to explore the requisites of a desirable defense in FL. Our study shows that (i) the batch size used in each training round affects the effectiveness of defenses in DL, (ii) the defenses investigated are somewhat effective and moderately influenced by batch size in FL settings and (iii) the non-IID data makes it more difficult to defend against data poisoning attacks in FL. Based on the findings, we discuss the key challenges and possible directions in defending against such attacks in FL. In addition, we propose detect and suppress the potential outliers(DSPO), a defense against data poisoning attacks in FL scenarios. Our results show that DSPO outperforms other defenses in several cases.

Traffic sign recognition (TSR) system is essential for autonomous vehicle and is vulnerable to security threats from adversarial attacks. The existing adversarial attacks for TSR are invasive and suffer from poor concealment and high computational complexity, and thus have low feasibility in real-world scenarios. This paper proposes a non-invasive modulated LED illumination-based adversarial attack scheme. By generating luminance flashes imperceptible to human eyes through fast intensity modulation of lighting such as LED streetlights and exploiting the rolling shutter mechanism of CMOS sensors of in-vehicle imaging system, the proposed attack scheme can successfully perform adversarial attacks on TSR system by implanting luminance information perturbations into the images acquired by autonomous vehicle and thus poisoning the image data fed into TSR system. Depending on the modulation frequency and pattern of LED illumination, the proposed attack scheme enables denial of service (DoS) attack that leads to traffic sign detection failure and escape attack that leads to traffic sign misclassification, with the advantages of superior concealment, low computational complexity and high practical feasibility. Experiments are conducted with two benchmark datasets (GTSDB and GTSRB) and two state-of-the-art models of TSR detection and TSR classification, YOLOv5m and Sill-Net respectively, in both the digital and physical world. Experimental results show that the proposed DoS attack on the TSR detection model (YOLOv5m) can reach the success rate of 90.00% and the proposed escape attack on the TSR classification model (Sill-Net) can achieve the success rate of 35.00%.

With the ubiquity of mobile devices and rapid development of cloud computing, mobile cloud computing (MCC) has been considered as an essential computation setting to support complicated, scalable and flexible mobile applications by overcoming the physical limitations of mobile devices with the aid of cloud. In the MCC setting, since many mobile applications (e.g., map apps) interacting with cloud server and application server need to perform computation with the private data of users, it is important to realize secure computation for MCC. In this article, we propose an efficient server-aided secure two-party computation (2PC) protocol for MCC. This is the first work that considers collusion between a malicious garbled circuit evaluator and a semi-honest server while ensuring privacy and correctness. Also, it can guarantee fairness when collusion does not exist. The security analysis shows that our protocol can securely compute any function f(x, y) against different types of adversaries in the malicious model. Also, the experimental performance analysis shows that this work outperforms the previous works for at least 10 times with the same security level.

How to track the target robustly is a challenging task in the field of computer vision. Occlusion as one of the most difficult problems, occurs due to the information lost when three-dimensional subjects are projected in two-dimensional interface, therefore, the 2D or 3D tracking algorithms which adopted depth information that expects to rely on three-dimensional special structure to resolve these problems and made somewhat progress. The 2D tracking algorithm is not efficient in fully using depth information, and the 3D tracking method is not robust because of the lack of mature 3D feature extraction method, which fairly restricts the actual tracking effect. Responding to above questions, we propose an adoption of adaptive quantified depth information, establish an adaptive hierarchical structure according to various scenarios. Hierarchical structure can filter the foreground and background information to reduce the interference in tracking, at the same time simplify the use of the depth information. Combined with kernel correlation filter tracking method, we design the algorithm using 2D apparent model under the spatial structures, which is efficient to deal with the problems of occlusion and the change of target scale, and prove its effectiveness on Princeton Tracking Dataset.

The Eclat algorithm is one of the most widely used frequent itemset mining methods. In the normal Eclat algorithm and its variants, it is inefficient to calculate the intersection size of itemsets by sequentially comparing elements, especially for large scale transactions. In this paper, we propose the fast Eclat algorithms that can quickly calculate the intersection size of multiple itemsets by using minwise hashing and the estimators. Minwise hashing is used to calculate the Jaccard similarity coefficient by mapping the elements of the sets to those of smaller sets. Two estimators are used to estimate the intersection size of itemsets based on the Jaccard similarity coefficient. Due to the “imperfect” hash function, minwise hashing may obtain a biased Jaccard similarity, which results in error between the real value and the estimated value of the intersection size. Thus, we proposed the HashEclat which uses the maximum of |A| and |B| to represent the union size |A U B|, and proposed the Sim-Eclat which uses the minimum of |A| and |B| to represent the intersection size |A fl B|. Furthermore, we use a boundary error E for better performance as follows: if E is large, the intersection size is determined by a traditional method, and the result is more accurate but takes longer to compute; otherwise, it will be the opposite. Both the theoretical analysis and experimental results show that the proposed algorithms can obtain almost all frequent itemsets with higher speed and less memory usage than other algorithms.

With the wide application of Internet of Things (IoT), a huge number of data are collected from IoT networks and are required to be processed, such as data mining. Although it is popular to outsource storage and computation to cloud, it may invade privacy of participants’ information. Cryptography‐based privacy‐preserving data mining has been proposed to protect the privacy of participating parties’ data for this process. However, it is still an open problem to handle with multiparticipant’s ciphertext computation and analysis. And these algorithms rely on the semihonest security model which requires all parties to follow the protocol rules. In this paper, we address the challenge of outsourcing ID3 decision tree algorithm in the malicious model. Particularly, to securely store and compute private data, the two‐participant symmetric homomorphic encryption supporting addition and multiplication is proposed. To keep from malicious behaviors of cloud computing server, the secure garbled circuits are adopted to propose the privacy‐preserving weight average protocol. Security and performance are analyzed.

Summary In this big data era, service providers tend to put the data in a third‐party cloud system. Social networking websites are typical examples. To protect the security and privacy of the data, data should be stored in encrypted form. This brings forth new challenges: how to allow different users to access only the authorized part of the data without decryption of the data. Attribute‐based encryption (ABE) offers fine‐grained access control policy over encrypted data such that users can decrypt successfully only if their attributes satisfy the policy. However, one drawback of ABE is that the computational cost grows linearly with the complexity of ciphertext policy or the number of attributes. The situation becomes worse for mobile devices with limited computing resources. To solve this problem, we adopt the offline/online technique combining with the verifiable outsourced computation technique to propose a new ciphertext‐policy ABE scheme using bilinear groups in prime order, supporting the offline/online key generation and encryption, as well as the verifiable outsourced decryption. As a result, most computations of key generation and encryption can be executed offline, and the majority of computational workload in decryption can be outsourced to third parties. The scheme is selectively chosen‐plaintext attack‐secure in the standard model. We also provide the proof of verifiability on outsourced decryption. The simulation results show that our proposed scheme can effectively reduce the computational cost imposed on resource‐constrained devices. Copyright © 2016 John Wiley & Sons, Ltd.

There is an increasing trend for data owners to store their data in a third-party cloud server and buy the service from the cloud server to provide information to other users. To ensure confidentiality, the data is usually encrypted. Therefore, an encrypted data searching scheme with privacy preserving is of paramount importance. Predicate encryption (PE) is one of the attractive solutions due to its attribute-hiding merit. However, as cloud is not always trusted, verifying the searched results is also crucial. Firstly, a generic construction of Publicly Verifiable Predicate Encryption (PVPE) scheme is proposed to provide verification for PE. We reduce the security of PVPE to the security of PE. However, from practical point of view, to decrease the communication overhead and computation overhead, an improved PVPE is proposed with the trade-off of a small probability of error.

In this paper, we propose a privacy-preserving reservation system for electric vehicles (EV) charging stations. Due to the short driving range of EV, frequent charging is necessary. A mechanism for charging station reservation for EV owners is desirable. Our proposed system allows the vehicle owner to reserve a number of charging stations along the intended route at different time-slots. Yet it is secure against misuse such that a user can only hold a limited number of reservations simultaneously. More importantly, our system can provide privacy for users. The charging station does not know the identity of the user who has reserved it. Thus location privacy can be protected. We demonstrate the practicality of our system with a prototype implementation on a smart phone. Finally, we also provide a security proof to show that our system is secure under well-known computational assumptions.

Nowadays wireless sensor network (WSN) is widely used in human-centric applications and environmental monitoring. Different institutes deploy their own WSNs for data collection and processing. It becomes a challenging problem when institutes collaborate to do data mining while intend to keep data privacy on each side. Privacy preserving data mining (PPDM) is used to solve the above problem, which enables multiple parties owning confidential data to run a data mining algorithm on their combined data, without revealing any unnecessary information to each other. However, due to the huge amount of data collected and the complexity of data mining algorithms, it is preferable to outsource most of the computations to the cloud. In this paper, we consider a scenario in which two parties with weak computational power need jointly run a k-means clustering protocol, at the same time outsource most of the computation of the protocol to the cloud. As a result, each party can have the correct result calculated by the data from both parties with most of the computation outsourced to the cloud. As for privacy, the data owned by one party should be kept confidential from both the other party and the cloud.

In document image processing, arc segmentation plays an important role in vectorization and graphic recognition. Moreover, the unsatisfactory results of several recent arc segmentation contests indicate that conventional methods are inadequate. This paper proposes a new arc segmentation algorithm called SymCAve (an acronym for Symmetry axis, Circle fitting and Average distribution points). First, we locate several seed points and adopt three strategies to ensure that the seed points are proper; then we calculate the center and radius utilizing the seed points. Second, the coordinates of the center and radius are adjusted by employing symmetry axes. Third, the average distribution points method is used to verify whether the points on the circumference are all black pixels. It is a complete circle if all of the points are black pixels. Otherwise, it is a partial circle if some of the points are black pixels and are continuous. Based on this information, the start and end angles of the partial circle can be determined. Finally, these arcs are verified to ensure that the results are accurate. Images and the evaluation tool were obtained from the GREC Workshop's Arc Segmentation contests, to test the systematic performance of the SymCAve algorithm. The experiments demonstrate that the proposed method can provide promising results. However, the algorithm has some drawbacks: it cannot detect a line with width of one pixel, small angles, and any large radius arcs. It is suited for segmenting images with appropriate symmetry axes.

The concept of legal professional privilege (LPP) in the Common Law is to enable a client to make full disclosure to his legal advisor for seeking advice without worrying that anything so disclosed will be used against him. Thus, some of the communications and documents between a legal advisor and his client can be excluded as evidence for prosecution. Protection of LPP information in the physical world is well addressed and proper procedures for handling LPP documents have been established. However, there does not exist a forensically sound procedure for protecting digital LPP information. In this correspondence, motivated by a real case of a commercial crime investigation, we introduce the LPP data integrity problem. While finding an ideal solution to solve the problem is difficult, we propose a practical solution that was adopted to solve the real case investigation. We also analyze the performance of our solution based on simulated data.

Verifying the integrity of a hard disk is an important concern in computer forensics, as the law enforcement party needs to confirm that the data inside the hard disk have not been modified during the investigation. A typical approach is to compute a single chained hash value of all sectors in a specific order. However, this technique loses the integrity of all other sectors even if only one of the sectors becomes a bad sector occasionally or is modified intentionally. In this paper we propose a k -dimensional hashing scheme, k D for short, to distribute sectors into a k D space, and to calculate multiple hash values for sectors in k dimensions as integrity evidence. Since the integrity of the sectors can be verified depending on any hash value calculated using the sectors, the probability to verify the integrity of unchanged sectors can be high even with bad/modified sectors in the hard disk. We show how to efficiently implement this k D hashing scheme such that the storage of hash values can be reduced while increasing the chance of an unaffected sector to be verified successfully. Experimental results of a 3D scheme show that both the time for computing the hash values and the storage for the hash values are reasonable.