As Deep Neural Networks (DNNs) become integral to critical applications, protecting their Intellectual Property (IP) has become paramount. Neural network watermarking is a technique that embeds unique identifiers into models, asserting ownership and deterring unauthorized use. However, sophisticated attacks can deactivate or remove these watermarks without significantly compromising model performance, undermining current protection strategies. In this article, we introduce the first method for reactivating deactivated neural network watermarks in altered DNN models without requiring access to the original model parameters or training data. By formulating the reactivation process as an optimization problem, we employ projected gradient descent to identify new trigger inputs that restore the embedded watermark. Regularization techniques are incorporated to ensure these triggers resemble legitimate inputs, enhancing both stealth and practicality. Through experiments on various benchmark datasets and model architectures, we demonstrate the effectiveness of our method against common model alterations, including fine-tuning, pruning, and surrogate model attacks. Our work addresses a critical gap in DNN IP protection, offering a robust and practical solution for watermark reactivation. This empowers model owners to assert their rights even in the face of advanced adversarial tactics.

The rapid advancement of deep neural networks, including large language models which has driven significant progress in biomedical data analysis and intelligent healthcare. However, strict privacy regulations and the decentralized nature of medical data make centralized model training impractical. One-shot Federated Learning (OFL) with knowledge distillation enables efficient cross-institutional collaboration on large models, but existing approaches often lack rigorous protection of confidentiality, authenticity, and model integrity. In this work, we introduce Bio-OFL, a secure and auditable one-shot federated learning framework for biomedical collaboration. BioOFL leverages CKKS homomorphic encryption and ECDSA digital signatures to provide end-to-end encrypted, authenticated uploads and introduces efficient encrypted-domain auditing to filter anomalous or adversarial contributions before aggregation. Formal security analysis and extensive experiments on MedMNIST datasets demonstrate that Bio-OFL delivers robust privacy and integrity guarantees for practical, large-scale biomedical AI, without compromising model utility.

In cloud and edge environments, distributed aggregation of multi-center electronic medical records and high-dimensional RNA sequencing features must comply with stringent privacy regulations (e.g., HIPAA, GDPR) and guarantee result integrity. Secure aggregation enables collaborative analyses—such as biomarker discovery and risk prediction—without exposing raw data; however, undetected computation errors or malicious tampering by the aggregator can undermine clinical decision support, regulatory compliance, and study reproducibility. We propose VSDA—Verifiable Secure Distributed Biomedical Data Aggregation—a framework that enhances traditional secure aggregation with discrete-logarithm-based signatures on both message vectors and secret keys. Participants can cryptographically verify, via signature-ciphertext consistency, that the aggregator has honestly executed the summation without revealing underlying data. VSDA supports fully parallelized operations with minimal overhead: each participant performs ($3 n+m-1$) scalar exponentiations and transmits ($m+1$) scalars, while the aggregator conducts ($m n+m$) exponentiations with no additional communication, achieving overhead reductions of several orders of magnitude compared to state-of-the-art methods. Under the discrete logarithm and one-time-pad assumptions, we provide formal security proofs for VSDA and empirically validate its efficiency, dropout resilience, and regulatory compliance on the TCGA RNA-Seq dataset. VSDA thus offers a practical, verifiable, and privacy-preserving solution for data integration and health information exchange in biomedical informatics.

In many fields, the need to securely collect and aggregate data from distributed systems is growing. However, designs that rely solely on encrypted data transmission make it difficult to trace malicious users. To address this challenge, we have enhanced the secure aggregation (SA) protocol proposed by Bell et al. (CCS 2020) by introducing verification features that ensure compliance with user inputs and encryption processes while preserving data privacy. We present LZKSA, a quantum-safe secure aggregation system with input verification. LZKSA employs seven zero-knowledge proof (ZKP) protocols based on the Ring Learning with Errors problem, specifically designed for secure aggregation. These protocols verify whether users have correctly used SA keys and their L∞, L2 norms and cosine similarity of data, meet specified constraints, to exclude malicious users from current and future aggregation processes. The specialized ZKPs we propose significantly enhance proof efficiency. In practical federated learning scenarios, our experimental evaluations demonstrate that the proof generation time for L∞ and L2 constraints is reduced to about 10-3 of that required by the current state-of-the-art method, RoFL (S&P 2023), and ACORN (USENIX 2023). For example, the proof generation/verification time of RoFL, ACORN and LZKSA for L∞ is 94s/29.9s, 78.7s/33.9s, and 0.02s/0.0062s for CIFAR10, respectively.

The intensified concerns arising from the widespread adoption of deep learning have led to increased scrutiny of intellectual property protection in DNN models. Existing audio watermarking techniques, predominantly based on traditional signal processing methods, struggle to balance robustness, imperceptibility, and defense resistance in the face of evolving adversarial attacks. These limitations underscore the urgent need for more effective watermarking solutions in the audio domain. In this paper, we propose a dynamic audio watermarking framework that introduces an optimization-based approach to attach robust and adaptable triggers at arbitrary positions within audio signals, and innovatively integrates boundary sample selection driven by forgetting events and an adaptive watermark trigger embedding technique based on the SNR. Comprehensive experimental results reveal that our scheme preserves high model performance while maintaining remarkable stealthiness and robustness, offering a secure and reliable solution for safeguarding intellectual property in the audio domain and advancing the field of DNN watermarking.

Federated learning (FL) is a promising approach for privacy-preserving machine learning, enabling collaborative model training across distributed devices without sharing raw data. However, FL faces significant challenges due to the nonindependent and identically distributed (non-IID) nature of data across devices, leading to difficulties in model convergence and generalization. In this paper, we propose FedDiT, a novel federated learning framework that combines knowledge distillation with vision transformers. FedDiT introduces the Distilled Vision Transformer (DTViT) model on the client side, incorporating a distillation token to enhance local learning and knowledge transfer. This approach significantly improves the robustness and performance of FL in non-IID environments. We validated FedDiT through extensive experiments on public datasets, and the results show that it outperforms existing FL methods in both accuracy and smoother convergence. Additionally, FedDiT achieves higher throughput compared to standard transformers and knowledge distillation methods, making it more efficient for practical deployment in federated learning scenarios.

Federated learning has found extensive application in the multimedia domain. However, due to its distributed nature, it is vulnerable to attacks such as Byzantine poisoning. To counteract malicious attacks, the secure aggregation process in federated learning requires input validation from participants. Existing input verification schemes, such as ACORN (USENIX Security 2023), ROFL (S&P 2023), et al., efficiently assess the validity of client inputs, but they fail to account for the impact of weights and do not support weighted secure aggregation. To address these issues, we propose α-SAV, an efficient weighted input verification scheme that utilizes Pedersen commitments to encrypt both privacy and weighted gradients. Our scheme incorporates a non-interactive zero-knowledge proof, the Sigma protocol, allowing clients to generate input proofs without interacting with the server. Verified inputs can then contribute to weighted aggregation. α-SAV is highly compatible, seamlessly integrating into existing federated learning frameworks with minimal additional cost. Experimental results demonstrate that the cost of α-SAV is linear. When trained on the MNIST dataset, the client computation time for α-SAV is 1.6 seconds, resulting in only 24% additional cost compared to ACORN and 3% compared to ROFL.

Federated learning, as an emerging distributed machine learning approach, enables collaborative model training while protecting data privacy. However, federated learning is vulnerable to Byzantine attacks and inference attacks. Existing solutions typically require semi-honest servers to perform secure aggregation or lack effective input validation mechanisms. To address these issues, we propose RIDE, a secure aggregation protocol for decentralized federated learning with input validation. RIDE utilizes pedersen commitments and efficient zero-knowledge proofs to verify whether model updates comply with predefined constraints, ensuring client input privacy and integrity. Additionally, RIDE employs a publicly verifiable secret sharing scheme, ensuring that only validated model updates are aggregated, even in the presence of malicious clients or client dropouts. Experimental results on four real datasets demonstrate the effectiveness of our solution. For example, RIDE has a maximum bandwidth overhead of 7.11MB, which is only 1.31× that of the most popular secure aggregation protocol (CCS 2020), and the computational cost of RIDE’s execution on the CIFAR-10 L dataset is 109.88s, which is 7.28× faster than the current state-of-the-art protocol RoFL (S&P 2023).

Knowledge Tracing (KT) plays a pivotal role in educational research by predicting students’ academic performance and enabling personalized learning interventions through dynamic knowledge state modeling. However, existing KT methods mainly focus on learning outcomes, neglecting the complexity of learning behaviors and underutilizing diverse features, which hinders scalability and interpretability. This study proposes a novel framework, Residual GRU-based Knowledge Tracing with Knowledge Attention (RGAKT), which predicts a student’s next response based on past interactions. The framework aims to enhance temporal dynamics modeling, explainability, and predictive accuracy in knowledge tracing tasks. RGAKT incorporates individual student differences and utilizes a two-layer GRU architecture with residual connections to improve temporal pattern recognition while reducing training complexity. Additionally, a knowledge attention (k-attention) mechanism is introduced to dynamically prioritize relevant interactions, making the model’s decision process more transparent and improving predictive accuracy. Extensive experiments on multiple benchmark datasets demonstrate that RGAKT outperforms state-of-the-art KT models, achieving a 4.48% increase in AUC, a 5.41% improvement in accuracy, a 19.17% reduction in RMSE, and a 3.77% decrease in MAE. These results highlight the model’s superior predictive capabilities and adaptability across diverse learning environments, establishing a new benchmark for KT performance. By integrating GRU-based temporal modeling with dynamic attention mechanisms, the RGAKT framework significantly advances the field of knowledge tracing, driving the development of personalized educational technologies and more adaptive learning systems.

Secure multi-party computation (MPC) allows multiple parties to jointly compute a task while preserving the privacy of their individual inputs. In recent years, researchers have extensively studied various security protocols to improve the efficiency of MPC and reduce communication overhead. However, the efficiency of MPC in an$n$-party setting still remains a challenge. In this paper, we introduce WINK, a MPC framework designed for the semi-honest security model that accommodates an arbitrary number of participants. We design an$n$-party multiplication protocol in the$n$-out-of-$n$arithmetic secret sharing setting and, based on this, construct independent multiplication triples. We leverage a RLWE-based oblivious linear evaluation (OLE) technique to batch generate multiplication triples between any two parties, significantly reducing communication overhead. Furthermore, we design a constant-round n-party comparison protocol based on the properties of odd-sized rings and garbled circuits. This protocol completes in only three rounds of communication with the help of additional non-colluding semi-honest servers, marking a significant improvement over previous$n$-party comparison protocols. We prove that our framework achieves universally composable security. Moreover, experimental results demonstrate that our proposed comparison protocol outperforms existing comparison protocols in terms of efficiency.

The Internet of Things (IoT) has become a transformative force across various sectors, including healthcare, offering new opportunities for automation and enhanced service delivery. The evolving architecture of the IoT presents significant challenges in establishing a comprehensive cyber‐physical framework. This paper reviews recent advancements in IoT‐driven healthcare automation, focussing on integrating technologies such as cloud computing, augmented reality and wearable devices. This work examines the IoT network architectures and platforms that support healthcare applications while addressing critical security and privacy issues, including specific threat models, attack classifications and security prerequisites relevant to the healthcare sector. This study highlights how emerging technologies like distributed intelligence, big data analytics and wearable devices are incorporated into healthcare to improve patient care and streamline medical operations. The findings reveal significant potential for IoT to transform healthcare practices, particularly in‐patient monitoring, and clinical decision‐making. However, security and privacy concerns continue to be a substantial barrier. The paper also explores the implications of global IoT and ehealth strategies and their influence on sustainable economic and community growth. It proposes an innovative cooperative security model to mitigate security risks in IoT‐enabled healthcare systems. Finally, it identifies key unresolved challenges and opportunities for future research in IoT‐based healthcare.

Split Learning (SL) is a distributed privacy-preserving learning methodology designed to address the challenges associated with the deployment of large-scale deep neural networks on medical devices, while simultaneously safeguarding the privacy of medical data. However, both the forward and backward propagation of the model require communication between the medical devices and high-performance servers, resulting in significant communication overhead and high latency. In this paper, to reduce the communication overhead from the client to the server during forward propagation, we propose an autoencoder layer based on attention mechanisms and triple compression. To reduce the communication overhead from the server to the client during backward propagation, we proposed an average loss threshold algorithm to decrease the frequency of client updates. Compared to the original Split Learning algorithm, after incorporating the method proposed in this paper, the communication overhead during forward propagation decreased by an average of 93%, and during backward propagation, it decreased by an average of 96%. The total communication overhead decreased by an average of 95%. The model’s accuracy loss was between 0% and 1%. In terms of communication compression, compared to the SOTA SL–BSL, the overall communication overhead was reduced by an average of 92%.

The integration of Artificial Intelligence (AI) in the healthcare sector has marked significant advancements, and Federated Learning (FL) has further facilitated the amalgamation of Federated Medical Imaging. However, this integration has also sparked concerns regarding data privacy. Incorporating Differential Privacy (DP) into gradients effectively mitigates privacy leaks but at the cost of impacting model accuracy. Current research delves into DP within FL, with a focus on strategies for privacy budget allocation and noise addition. Nevertheless, the dynamic privacy requirements and resource optimization for actual medical applications are often overlooked, leading to resource wastage. This study introduces an innovative algorithm based on gradient component for adaptive noise scale optimization and privacy budget allocation, thereby enhancing privacy management while maintaining model accuracy. Our findings reveal that, compared to traditional DP techniques, our approach achieves an average accuracy improvement of 3.47% in RSNA-ICH Acc and an enhancement of up to 172.33% in Dice score for Prostate MRI Dice under various privacy budgets, demonstrating the substantial efficacy of our method in the domain of federated medical imaging.

With the rapid proliferation of social networks, individuals now have greater access to news with increased speed. Simultaneously, there has been a heightened emphasis on detecting and mitigating the dissemination of fake news. One notable limitation of existing fake news detection models is their inability to effectively integrate multi-modal features, as they typically only establish connections between unimodal features, neglecting the potential synergies and complementarity among different modes. To address this issue, we introduce the Multi-modal Adaptive Co-attention fusion Contrastive learning Network (MACCN) for enhancing the detection of fake news by improving the fusion of textual and visual features. Our approach commences by employing distinct encoders to construct a high-level feature space for each modality. Subsequently, the Adaptive Co-attention Fusion Network is employed to establish strong correlations between textual and visual features, leading to a comprehensive representation. Ultimately, the model's performance is further enhanced through the application of contrastive learning, resulting in a more precise detection of fake news. We conducted an extensive series of experiments on three diverse datasets, and the results conclusively demonstrate that MACCN adeptly captures the interplay between multi-modal features, surpassing the performance of state-of-the-art methods.

As the Internet of Things (IoT) continues to evolve, the ability to share data has become an integral aspect of cloud computing services.Nevertheless, the persistent issue of data security presents a formidable challenge within the domain.This study introduces a novel blockchain oriented data sharing architecture designed to reinforce data security while optimizing efficiency.The architecture is structured around advanced smart contracts and security mechanisms that register cloud-based data activities on a blockchain ledger.In instances of anomalous activities, the blockchain is scrutinized by a centralized cloud service to identify and hold accountable any malicious gateways.The framework employs robust authentication and secure data transmission protocols to fortify data security.Furthermore, it utilizes sophisticated yet efficient partial decryption algorithms within smart contracts to alleviate the computational load on end users.Blockchain's capability to provide traceable historical records underpins the system's ability to meet stringent data safety standards through transparent and open oversight.Empirical evidence underscores the effectiveness of the proposed system in safeguarding data exchanges across various clients while maintaining high operational efficiency.

Visual Question Answering (VQA) as a cutting-edge domain blending computer vision and natural language processing has garnered significant research momentum. Nev-ertheless, latest VQA frameworks which leverage CNN and RNN technologies to extract features and delve into the multimodal interplay between text and images, encounter difficulties in integrating local features with global dependencies. This integration challenge hinders the models' ability to precisely grasp the pivotal aspects of the answer. Furthermore, the dearth of training data poses a significant impediment to models' effective learning of multimodal information exchange and problem semantics. To address these issues, our model introduces a new method called CMACC (Cross-Modal Adversarial Contrastive Learning Based on Co-Attention). Through cross-modal adversarial contrastive learning, combined with common attention, image and text information are integrated. Adversar-ial learning aligns the latent feature distribution between text and images, and contrastive learning aligns multimodal sample features in the same context to enhance the adaptability of the model. In addition, advanced data augmentation techniques are integrated to further enhance the model's adaptability to different scenarios and problem types. We conducted experimental evaluations on three widely used VQA datasets (VQA v1.0, VQA v2.0, and COCO-QA) and the results showed that CMACC has significant improvements in accuracy and generalization performance compared to traditional methods.

Federated Learning (FL), a distributed learning paradigm optimizing communication costs and enhancing privacy by uploading gradients instead of raw data, now confronts security challenges. It is particularly vulnerable to Byzantine poisoning attacks and potential privacy breaches via inference attacks. While homomorphic encryption and secure multi-party computation have been employed to design robust FL mechanisms, these predominantly rely on Euclidean distance or median-based metrics and often fall short in comprehensively defending against advanced poisoning attacks, such as adaptive attacks. Addressing this issue, our study introduces “Split-Aggregation", a lightweight privacy-preserving FL solution capable of withstanding adaptive attacks. This method maintains a computational complexity ofO(dkN+k3) and a communication overhead ofO(dN), performing comparably to FedAvg whenk= 10. Here,drepresents the gradient dimension,Nthe number of users, andkthe rank chosen during random singular value decomposition. Additionally, we utilize adaptive weight coefficients to mitigate gradient descent issues in honest users caused by non-independent and identically distributed (Non-IID) data. The proposed method’s security and robustness are theoretically proven, with its complexity thoroughly analyzed. Experimental results demonstrate that atk= 10, this method surpasses the top-1 accuracy of current state-of-the-art robust privacy-preserving FL approaches. Moreover, opting for a smallerksignificantly boosts efficiency with only marginal compromises in accuracy.

Document Similarity Detection (DSD) is significant in our real life applications. However, the existing methods ignore the privacy of what is contained in the documents uploaded on remote servers, thus reducing the applicability of these methods. The proposed scheme allows documents to be compared without revealing to those remote servers. For each document, the fingerprint set is calculated. The inverted index is constructed on the basis of the whole fingerprint set. The inverted index is widely used for efficient retrieval. This index is under protection by Paillier cryptosystem before it gets uploaded to the server.

The proper use of evolutionary operators is crucial to find optimal solutions in a search space. Moreover, the diversity of the population affects the performance of Evolutionary Algorithms (EAs). This article introduces an EA called BWEAD which balances the influence of the operators. The proposal also performs a statistical analysis of the population when the diversity is low and decides which solutions might be replaced. Then BWEAD is able to explore the search space and exploit the prominent regions. The BWEAD has been tested over the CEC2014 set of benchmark functions. The experiments provide competitive results showing an improvement of 30% in 30-dimensional and 50-dimensional functions in comparison with state-of-the-art algorithms, overcoming some addressed instances and providing evidence of its capabilities on complex optimization problems.

This paper proposes an alternative multilevel thresholding (MLT) image segmentation method by improving the behavior of the grasshopper optimization algorithm (GOA). This is achieved by using the operators of the sine-cosine algorithm (SCA) to work in a competitive manner with the operators of traditional GOA. This will lead to enhance the quality of the solutions during the updating process that will affect the convergence of the proposed GOASCA towards the global solution. In addition, the proposed GOASCA aims to minimize the difference between the fuzzy entropy and its opposite fuzzy entropy that is used as a fitness function to evaluate the quality of the solution. This objective function gives the GOASCA to explore the whole search space. To assess the quality of the obtained threshold values by GOASCA, a set of eight images are used which have different characteristics. Moreover, the results of GOASCA are compared with a set of well-known MLT image segmentation approaches, and these results have shown the high quality of GOASCA to segmented the image, as well as, shown that the current objective function provides results better than the traditional fuzzy entropy in terms of the performance measures of image segmentation.

Automatic clustering based hybrid metaheuristic algorithms has attracted the center of interest of scientists and engineers which become a hot topic for different data analysis applications. For example, image clustering, bioinformatics, image segmentation, and natural language processing. Where the process of determining the number and position of centroids is an NP-hard problem. So, this paper presents an alternative automatic clustering algorithm based on the hybrid between the atom search optimization (ASO) and the sine-cosine algorithm (SCA). The main objective of the proposed clustering method, called ASOSCA, is to find automatically the optimal number of centroids and their positions in order to minimize the CS-index (which refers to Compact-separated index). To achieve this goal, the ASOSCA uses SCA as a local search operator to improve the quality of ASO. The performance of the proposed hybrid method is compared with other metaheuristic methods; in which all of them are tested on sixteen clustering datasets and using different cluster validity indexes as Dunn, Silihouette, Davies Bouldin, and Calinski Harabasz. The experimental results show that the ASOSCA depict high superiority in comparison with other types of hybrid metaheuristic in terms of clustering measures.

A previous work proposed a method which can change a predicate encryption supporting inner product (IPE) scheme into a public key encryption with conjunctive keyword search (PECK) or public key encryption with disjunctive keyword search (PEDK) scheme. However, there are two problems in this method. The one is that the PEDK scheme based on this method has low efficiency on the time and space complexity. The other is that the PECK scheme and the PEDK scheme generated by using this approach can not be combine into one scheme which can support both conjunctive and disjunctive keyword search over encrypted data. To mitigate these concerns, we propose a method for constructing a scheme called public key encryption with conjunctive and disjunctive keyword search (PECDK), and give an instance. The comparison shows that our scheme can solve two problems mentioned above efficiently.

In this paper, we revisit the Property Matching problem and present a better indexing scheme for the problem. Let T be a text of length n with property p, and P be a pattern of length m, both strings are over a fixed finite alphabet. In particular, the existing data structures all require O(n log n)-bit space, where n is the length of the text. By using compressed suffix array and other supporting data structures, we propose a new index structure for the problem. We discuss the index structure and searching process for the case |π| = O(n/ log n) and |π| = Ω(n/ log n). Our index only needs nHk(T)+O(n log |Σ|)-bits and nHk(T)+nH0(A)+O(n(log |Σ|+log log n)) bits space for the above cases respectively (A is an array with length n here), while needing a little more searching time as return.

In this paper we describe a new approach to distinguish and extract text from images with various objects and complex backgrounds. The goal of our approach is to present characters in images with clear background and without other objects. The proposed approach mainly includes two steps. Firstly, a density-based clustering method is employed to segment candidate characters by integrating spatial connectivity and color feature of characterspsila pixels. In most images, colors of pixels in one character are commonly non-uniform due to the noise. So a new histogram segmentation method is proposed in this step to obtain the color thresholds of characters. Secondly, priori knowledge and texture-based method are performed on the candidate characters to filter the non-characters. Experimental results show that the proposed approach has a good performance in character extraction rate.