Postquantum key-encapsulation mechanisms (KEMs) usually use the well-known Fujisaki-Okamoto (FO) transformation during key decapsulation to achieve chosen ciphertext attack (CCA) security. In the FO transformation, the re-encryption procedure depends on the message. The side-channel leakage of re-encryption can be exploited to recover the coefficients of the secret key under chosen ciphertexts, even if the KEM scheme is CCA-secure. However, it still requires a large amount of trace during the profiling and attacking phase. In this work, we introduce the first nonprofiled deep learning-based attack (NPDLA) on lattice-based KEMs. We propose a chosen ciphertext method that is suitable for NPDLAs. The horizontal chunking strategy is employed to partition the coefficients into chunks, enabling the independent recovery of multiple secret key coefficients within each chunk. Then, we adopt an incremental learning strategy to allow the deep learning model to gradually learn the knowledge of each chunk. Moreover, we push the limits of traditional NPDLAs by combining the unsupervised domain adaptation with the correlation distinguisher, eliminating the necessity of neural network training for each possible key guess. The feasibility of the attack is verified by practical experiments for the unprotected and masked implementations of Kyber on the ARM Cortex-M4.

The Internet of Things (IoT) is a key engine of global socio-economic transformation, where data sharing stands as a central catalyst for the IoT market's growth. However, data security and privacy concerns significantly impede the advancement of IoT data sharing. Consequently, Attribute-Based Encryption (ABE), offering fine-grained access control, is increasingly favored by data users. Unfortunately, existing ABE schemes still face these drawbacks: (1) the encryption and decryption computation overhead grows linearly with attributes; (2) keyword searches within ciphertexts are intricate and inefficient; (3) the access policy is at higher risk of privacy disclosure. To address these issues, this paper presents an efficient searchable and verifiable scheme with partial hidden policy for IoT (ESVPH). This scheme not only provides flexible keyword-based search and re-encryption verification, but also achieves fixed costs for encryption, decryption, searching and verifying. Additionally, ESVPH introduces an access policy where attribute names are disclosed while their values remain concealed, thereby enhancing user privacy. In conclusion, the scheme offers outstanding performance in computation and communication, proving its feasibility for practical IoT data sharing through rigorous proofs and extensive experimentation.

Blockchain has become a superior carrier for data storage due to its decentralization and immutability. How to realize privacy-preserving data sharing and controlled transfer of blockchain data access rights has become a challenging issue. Autonomous path proxy re-encryption (PRE) allows the delegator to designate a series of delegatees to obtain decryption privileges according to a predefined sequence in a multihop manner. However, it faces the hop-skipping issue, where a malicious proxy can directly re-encrypt the ciphertext to the delegatees several hops afterward, by skipping one or more delegatees in between, destroying the decryption privileges of the skipped delegatees. Furthermore, the intrinsic nature of PRE makes it vulnerable to a collusion attack, wherein the proxy and a delegatee might collude to construct a decryption device that can decrypt all the delegator's ciphertexts, posing a serious threat to the delegator's privacy. To address these challenging issues, we propose a generalized collusion and hop-skipping resistant autonomous path proxy re-encryption (CHRAP-PRE). First, we decentralize the proxy's privileges of re-encrypting ciphertexts to resist a collusion attack. Second, we carefully design the decryption token mechanism so that only the person who gets the correct token can do the decryption, which is authorized by all previous persons in the path, thus controlling the decryption privileges to solve the hop-skipping problem. Finally, we formally prove that our proposed CHRAP-PRE achieves IND-HRA security under the decisional bilinear Diffie-Hellman (DBDH) assumption, resisting both collusion and hop-skipping attacks. Our comprehensive performance evaluation demonstrates that our scheme offers enhanced security while reducing communication overhead compared to the state-of-the-art.

Graph Neural Networks (GNNs) have powerful representation capabilities for graph data, achieving excellent performance across various fields. Considering the scarcity of labels in real-world scenarios, graph self-supervised learning (GSSL) has gained increasing attention due to its ability to train without relying on labels. However, recent studies have revealed that GNNs are vulnerable to stealthy backdoor attacks in GSSL scenarios, enabling the encoder to learn backdoor features simply by injecting triggers. Existing graph backdoor defense methods mainly focus on supervised settings and cannot be directly transferred to self-supervised scenarios due to the lack of label guidance. To bridge this gap, we propose GDetox, the first backdoor defense approach against backdoored encoders in GSSL. GDetox aims to eliminate backdoor logic in encoders while maintaining the encoder's original performance. Specifically, GDetox can purify the graph backdoor encoder based on the self-supervised distillation approach without relying on label information. Further, we introduce an adversarial contrastive learning that augments node representations without relying on labels to enhance teacher model performance, thereby improving distilled encoder performance. We evaluate the defense performance of GDetox on four node classifications and four graph classification datasets by comparing with four state-of-the-art (SOTA) defense methods against seven latest backdoor attack methods on GSSL. Extensive experiments demonstrate that GDetox far outperforms the SOTA defense methods, reducing the attack success rate to 4% with negligible degradation in encoder performance (within 2%) in both node-level and graph-level tasks.

Deep learning-based recommendation systems are highly vulnerable to data poisoning attacks, where adversaries manipulate user interactions to degrade model integrity. We hypothesize that combining an active robust loss with a passive GAN-based detection will significantly reduce poisoning impact in recommendation systems without sacrificing utility. We propose a robust and adaptive dual-defense framework: the active defense integrates a crafted loss function to mitigate poisoning effects while maintaining model performance. The passive defense employs a Generative Adversarial Network (GAN)-based detection model to identify and filter poisoned data, enhancing detection accuracy and system security. The framework supports classical matrix factorization (MF) model and large language model (LLM)-based pipelines and scales to large datasets. Extensive experiments across multiple real-world datasets at varying poison rates show that our method outperforms representative defenses, consistently reducing attack success without sacrificing recommendation quality. The framework also admits a federated instantiation, where robust training and GAN-based detection run on clients and only privacy-preserving summaries are aggregated. The proposed method significantly improves the robustness and adaptability of recommendation systems under data poisoning attacks.

Reinforcement learning (RL) has been increasingly adopted in IoT systems for tasks such as resource allocation and control. However, in privacy-critical and resource-constrained environments, existing privacy-preserving RL schemes suffer from high computational cost, slow convergence, and limited scalability due to the use of homomorphic encryption (HE) or differential privacy (DP). We propose a distributed privacy-preserving Q-learning framework that enables secure and efficient policy updates across multiple clients. Each client independently trains a local Q-table and encrypts it using a sparse matrix transformation combined with additive secret sharing of structured perturbations. The encrypted Q-tables are uploaded to a cloud server for aggregation and averaging without decryption. The encrypted global Q-table is then returned and decrypted locally using the inverse sparse matrix. Experimental results on four benchmark environments (CartPole-v1, MountainCar-v0, Acrobot-v1, and LunarLander-v3) show that our scheme achieves up to 92% reduction in computation time compared to the FHE-based method, while maintaining comparable reward performance and faster convergence.

Provenance graph-based anomaly detection, particularly for Advanced Persistent Threat (APT) detection, addresses the issues of large-scale graphs and data imbalance. However, existing methods struggle with information loss, high computational complexity, and low detection accuracy. To address the above challenges, this paper proposes TraceCluster, a lightweight and adaptive clustering-based Subgraph Attention Network (SAN) for APT detection in provenance graph. TraceCluster mitigates the neighborhood explosion problem by clustering nodes to partition large-scale graphs, thus reducing reliance on the global graph while preserving local neighborhood information. Furthermore, the method dynamically models complex inter-node dependencies within subgraphs. It employs an attention mechanism to adaptively highlight the most relevant connections. This enhances node representations and improves overall feature extraction. This design substantially reduces memory consumption and avoids the high computational complexity of global graph processing. In addition, an adaptive category-weighting loss function assigns variable weights to different classes, improving the detection of rare and anomalous behaviors. Experimental results show that on the OpTC dataset, the currently faster method is 37-fold and 3-fold slower than our approach in terms of inference time respectively. Furthermore, in the nine real-world scenarios of four evaluated datasets, TraceCluster outperforms state-of-the-art (SOTA) approaches in terms of overall performance, especially in node-level APT detection tasks.

Deep learning-based digital watermarking technology plays a crucial role in copyright protection and has been extensively researched. Current CNN-based robust watermarking methods embed the watermark into the deep features of the image through convolution operations. Despite the success of CNN-based watermarking frameworks, their reliance on capturing features at fixed spatial positions makes it challenging to address desynchronization caused by geometric distortions. To solve the above problem, we propose a robust watermarking framework based on the Swin Transformer named GResMark. Intuitively, GResMark leverages the spatial self-attention mechanism of the Swin Transformer to capture global and long-range dependencies among image tokens. Since geometric attacks preserve these relationships, GResMark achieves improved adaptability and robustness against such distortions. Specifically, we design the Locally-enhanced Channel Attention (LeCA) Swin Transformer block and Frequency Channel Attention (FCA) Transformer block as the backbone of our framework, enabling more effective dependency modeling for robust watermarking. In addition, to mitigate the impact of invalid pixel values in the watermarked images generated during training, which exceed the valid range defined by the underlying data format, we introduce a constrained loss that enhances the quality of watermarked images. Experimental results demonstrate that GResMark outperforms existing state-of-the-art (SOTA) watermarking methods in terms of robustness to geometric distortions and watermark capacity. For geometric attacks, GResMark achieves an extraction accuracy exceeding 98 % while maintaining superior visual quality.

With the widespread deployment of smart sensors, multi-view data has been widely used. Accordingly, multi-view processing algorithms are increasingly researched, among which the cluster-weighted kernel k-means method is an effective approach to dig up information of different views. However, large-scale multi-view data make it difficult to conduct processing algorithms. Therefore, outsourcing complex computations to servers based on privacy-preserving techniques is an effective solution that enables efficient multi-view data analysis. In previous secure outsourcing schemes, the efficiency of the outsourcing process and the fairness of outsourcing transactions are still challenging issues that have not been addressed. In this paper, we propose a blockchain-aided secure and fair multi-view data outsourcing computation scheme. We present an efficient matrix encryption method utilizing a novel secret key matrix to complete cluster-weighted kernel k-means algorithm securely. Different from previous works, we first apply the sparse symmetric orthogonal matrix to encrypt and decrypt sensitive data matrices, which avoids inverse or transposed secret key matrix computation and enhances the efficiency of the outsourcing process. Additionally, we introduce smart contracts to achieve fair outsourcing transactions aided by blockchain. We verify the returned result with the assistance of verifiers based on encrypted data, which improves the efficiency and security of outsourcing transactions. The experimental results indicate that our scheme is 4.72% to 8.52% superior to the state-of-the-art matrix outsourcing computation schemes and achieves 55.79% to 91.95% efficiency improvement compared to the original multi-view data processing method.

Fully homomorphic encryption (FHE) supports computing over encrypted data without requiring decryption, promising the privacy and confidentiality of data. With the rapid development of the Internet of Things (IoT), the SEAL-Embedded library is developed to implement CKKS on resource-constrained embedded devices efficiently. Although FHE is secure against mathematical cryptanalysis, the implementations on embedded devices are vulnerable to physical attacks. In this work, we propose a novel side-channel attack on the SEAL-Embedded library. We analyze the vulnerabilities in the rejection sampling of the SEAL-embedded library. The bit concatenation used for memory compression leads to multivariate leakage of key coefficients, and there is a correlation between the leakages of different coefficients. To enhance the accuracy of coefficient recovery, we design a deep learning model that adapts to the leakage features arising from the bit concatenation implementation for memory compression. Additionally, we construct the factor graph of the rejection sampling procedure to jointly exploit three sources of leakage: the sampled key coefficients, the random bytes before modulo reduction, and the Hamming weight of the concatenated bits, thereby improving the attack success rate on key coefficients. The proposed attacks are evaluated on the ARM Cortex-M4, which is widely used in IoT applications. The experimental results demonstrate that the proposed method achieves a 99% attack success rate in the single-trace attack, representing a 13.4% improvement over the widely adopted CNN-based attack.

Deep learning-based watermarking has made remarkable progress in recent years. To achieve robustness against various distortions, current methods commonly adopt a training strategy where a \underline{\textbf{s}}ingle \underline{\textbf{r}}andom \underline{\textbf{d}}istortion (SRD) is chosen as the noise layer in each training batch. However, the SRD strategy treats distortions independently within each batch, neglecting the inherent relationships among different types of distortions and causing optimization conflicts across batches.As a result, the robustness and generalizability of the watermarking model are limited. To address this issue, we propose a novel training strategy that enhances robustness and generalization via \underline{\textbf{meta}}-learning with \underline{\textbf{f}}eature \underline{\textbf{c}}onsistency (Meta-FC). Specifically, we randomly sample multiple distortions from the noise pool to construct a meta-training task, while holding out one distortion as a simulated ``unknown'' distortion for the meta-testing phase.Through meta-learning, the model is encouraged to identify and utilize neurons that exhibit stable activations across different types of distortions, mitigating the optimization conflicts caused by the random sampling of diverse distortions in each batch.To further promote the transformation of stable activations into distortion-invariant representations, we introduce a feature consistency loss that constrains the decoded features of the same image subjected to different distortions to remain consistent.Extensive experiments demonstrate that, compared to the SRD training strategy, Meta-FC improves the robustness and generalization of various watermarking models by an average of 1.59\%, 4.71\%, and 2.38\% under high-intensity, combined, and unknown distortions.

Machine Learning as a Service (MLaaS) provides robust solutions for deploying deep learning inference in cloud environments. However, it also raises serious privacy concerns regarding user data and proprietary model parameters. Numerous hybrid cryptographic protocols that integrate homomorphic encryption (HE) and garbled circuits (GC) have been proposed to enable secure inference with low latency. In these protocols, the homomorphic evaluation of linear operations remains the primary performance bottleneck and warrants further optimization. In this work, we propose novel optimizations for HE-based linear computations within the hybrid cryptographic framework for secure neural network inference. Specifically, we devise two efficient strategies for homomorphic matrix-vector multiplication and convolution. For matrix-vector multiplication, we introduce a grouped diagonal extraction technique that encodes the weight matrix more compactly and enables configurable ciphertext rotation reuse, while for homomorphic convolution, we present a group-wise combine-and-merge evaluation method. Both methods significantly reduce the number of required ciphertext rotations. Our approach achieves up to a 3.9x speedup in matrix-vector multiplication and a 2.9x improvement in convolution over state-of-the-art (SOTA) solutions. The HE-GC hybrid secure convolutional neural networks (CNN) inference framework incorporating these enhancements yields speedups of 2.5x on widely used ResNets deep learning architectures.

The demand for mobile terminals to participate in data services is increasingly vital. The General Data Protection Regulation (GDPR) has established several principled requirements for data services. Existing studies focusing on data service put emphasis on data privacy and accessibility. However, they face challenges in achieving data forgetability and portability on mobile devices under GDPR and lack consideration of usage control. In this article, we propose ADSS, an app-level data service scheme for mobile devices that can be available-but-invisible and guarantee fine-grained usage control. ADSS addresses the challenges by executing the logic of data usage in the Trusted Execution Environment (TEE) and managing the TEE states (i.e., data usage states) in the blockchain smart contracts. It not only satisfies the requirements of GDPR, ensuring strong security and confidentiality guarantees, but also enables the functionality of "pay-per-use". We implement a prototype of the ADSS framework based on ARM Trustzone and conduct experimental evaluations. The results demonstrate that our scheme brings high efficiency compared with other data service schemes and exhibits feasibility on mobile-grade devices.

Imbalanced clustering algorithm plays a vital role in fields, such as fault detection in finance, network security and medical diagnosis. The Imbalanced Clustering with Theoretical Learning Bounds (ICTLB) algorithm is a novel imbalanced clustering algorithm but could incur high computational costs due to extensive matrix operations, making it less practical for resource-limited devices. Outsourcing computations to cloud servers can alleviate client burdens but need to solve data privacy issues and result verification problem. In this paper, we propose an efficient, secure, and verifiable outsourcing scheme for the ICTLB imbalanced clustering algorithm. We design a novel encryption method based on sparse matrices and random permutations, which effectively protects the privacy of the input data while ensuring minimal computational overhead on the client side. Our scheme also integrates a robust verification mechanism, allowing the client to validate the correctness of results returned by the cloud server. Experiments show that the proposed scheme can improve efficiency by 28.88% to 52.48% comparable to the original ICTLB algorithm across various datasets.

Graph Neural Networks (GNNs) are highly susceptible to numerous adversarial attacks, among which the backdoor attack is one of the toughest to deal with due to the fact that it can lead to misclassification of the model. Similar to Deep Neural Networks (DNNs), backdoor attacks in GNNs work by an attacker changing a portion of the graph data with a hidden trigger and modifying their labels to target labels, which induces the model to learn the trigger feature during its training phase. Although recent defense techniques have emerged, approaches based on explainability and data isolation often fail to detect malicious samples with covert triggers, while discrepancy learning methods tend to degrade performance by removing useful features. To overcome these limitations, we propose a novel backdoor defense method, named GraphCleanse, on GNNs that can effectively eliminate the possible backdoor features during the training process. Specifically, GraphCleanse can easily break the strong correlation between backdoor features and target labels based on graph contrastive training. To further improve the model accuracy, we present a mutual information maximization method to learn the important feature information in the labeled credible samples and unlabeled suspicious samples by clustering the features obtained from the graph contrastive encoder. Compared with the potential solutions, such as randomized smoothing, GraphCleanse effectively avoids the negative influence of backdoored samples while maintaining a high model performance. Extensive experimental evaluations on four benchmark datasets demonstrate that GraphCleanse can reduce the attack success rate to 10% with less performance degradation (within 7%).

Homomorphic permutation is fundamental to privacy-preserving computations based on batch-encoding homomorphic encryption. It underpins nearly all homomorphic matrix operations and predominantly influences their complexity. Permutation decomposition as a potential approach to optimize this critical component remains underexplored. In this paper, we propose novel decomposition techniques to optimize homomorphic permutations, advancing homomorphic encryption-based privacy-preserving computations. We start by defining an ideal decomposition form for permutations and propose an algorithm searching for depth-1 ideal decompositions. Based on this, we prove the full-depth ideal decomposability of permutations used in specific homomorphic matrix transposition (HMT) and multiplication (HMM) algorithms, allowing them to achieve asymptotic improvement in speed and rotation key reduction. As a demonstration of applicability, substituting the HMM components in the best-known inference framework of encrypted neural networks with our enhanced version shows up to a 3.9x reduction in latency. We further devise a new method for computing arbitrary homomorphic permutations, specifically those with weak structures that cannot be ideally decomposed. We design a network structure that deviates from the conventional scope of decomposition and outperforms the state-of-the-art technique under a limited rotation key budget, achieving a speed-up of up to 1.69x.

The blockchain technology provides a revolutionary solution for information exchange through its decentralized, tamper-proof, and highly secure characteristics. It has wide application in many industries, with the potential to improve efficiency, reduce costs, and promote innovation. However, the full replication mechanism of blockchain results in the need for each device to store complete blockchain data, leading to inefficient storage. Additionally, as the scale of the blockchain network expands, the increasing data volume and frequent transactions can cause network congestion and latency, posing scalability issues for blockchain. Coded sharding blockchain has been proposed to address these issues. However, the current solutions face challenges such as dealing with malicious nodes and low computational efficiency, which hinder the enhancement of their scalability and computational performance. To resolve these problems, we propose AdaptiveShard by combining coded sharding blockchain with adaptive verifiable coded computing (AVCC). This solution is designed based on the Unspent Transaction Output (UTXO) model and is suitable for cryptocurrency transaction scenarios. Compared to traditional coded sharding blockchain solutions, AdaptiveShard can: 1) enhance the computational performance of coded sharding blockchain during block validation by combining AVCC with Gaussian variant of Freivalds algorithm (GVFA), reducing the decoding complexity to O(N(2)logN) ; 2) validate the computation results of each shard using GVFA and replace balance check verification functions with matrix multiplication, reducing the computational complexity of verification to O(root n) ; 3) reduce the additional number of nodes required to resolve malicious nodes from two to one using verifiable computation; 4) balance the system in the presence of straggler or malicious nodes through dynamic coding techniques, eliminating their impact and improving system reliability. Experiments demonstrate that at t=1000, the throughput is 25.6% higher compared to Polyshard. Compared to the solution without dynamic coding, the solution with dynamic coding can reduce the running time by 9.7% at t=50.

With the rapid advancement of technology, cloud computing has emerged as the most popular and promising service platform. A cloud user can delegate heavy computation tasks to cloud servers. To ensure the correctness of outsourced processing (e.g., machine learning and data mining), the cloud server must prove that the processing has been executed properly. However, even without malicious intent, it is possible for a cloud server to produce incorrect results. Consequently, clients may outsource the same task to multiple cloud servers and receive various results, aiding them in selecting the best outcome. To protect data privacy, the cloud server must encrypt the results before sending them back to the user. Yet, processing and verifying encrypted results remain significant challenges. To avoid the expensive computational overhead of decrypting ciphertexts from cloud servers one by one, clients prefer to use homomorphic encryption (HE) to obtain the combined output from a single server. However, existing schemes fall short of efficiently verifying the correctness of computations over encrypted data processed by multiple cloud servers, especially in extracting the results computed by each server. In this paper, we introduce a new framework for verifiable outsourced computing systems. In this system, each cloud server's computation result is protected by Paillier encryption, and the edge server can verify these results using zero-knowledge proofs and aggregate the verified ciphertexts. The client can extract the combined plaintext through the Base-3 conversion algorithm to identify each cloud server's results and any non-participating servers. We also prove the security of our scheme and analyze its performance from both theoretical and experimental aspects. Performance analysis shows that our system significantly reduces the client's workload and is userfriendly

For the huge amount of data from the Internet of Things (IoT) devices, multiple kernel learning is a widely concerned issue in data analyzing, among which the multiple kernel fuzzy clustering (MKFC) algorithm is an effective approach for extracting linear features in high-dimensional space. For a time-consuming multiple kernel clustering task, it is meaningful to find a secure and efficient outsourcing scheme in the edge-end collaborative architecture, which utilizes edge computing resources while resisting untrusted edge servers. However, existing secure outsourcing schemes cannot align well with the distributed and real-time characteristics of edge computing due to their complex encryption processes. In this article, we propose a secure MKFC outsourcing scheme based on a novel matrix blinding method. The proposed novel matrix blinding method conducts two related encryption operations with disturbance terms, which avoids specific disturbance elimination computations, to reduce the computational burdens in the decryption phase. Additionally, we introduce a sampling verification method to detect the server's deceptive behaviors. The theoretical analysis demonstrates that our scheme guarantees data privacy and has the capability to verify incorrect results. The experimental results indicate that our scheme is 6.73% superior to other schemes on average when conducting matrix outsourcing computation and enhances the efficiency of conducting the MKFC algorithm by 10.44% to 55.70% on different datasets.

Deploying Federated Learning within blockchain is widely adopted due to its ability to leverage decentralized architecture to ensure security and privacy while facilitating the allocation of training rewards. However, in real-world scenarios, an unreasonable incentive mechanism may lead to clients' reluctance to participate in training due to insufficient or unfair reward distribution. Furthermore, clients might adopt abnormal training strategies that negatively impact the final model performance for reduced costs or additional gain. Thus, designing a fair and robust incentive model that promotes continuous and honest participation remains a significant challenge. In this paper, we propose a game-theoretic dynamic incentive model termed FRFL, which allocates rewards based on client contributions and reputation. The model incentivizes honest participation by introducing a game theory-based reward distribution strategy. By analyzing payoffs for bounded rationality clients under various conditions, we calculate the Sequential Equilibrium and the Evolutionarily Stable Strategy (ESS) to assess model stability and demonstrate through simulations that our model successfully converts participants employing unconventional strategies into honest participants, thereby ensuring system robustness.