With the widespread integration of large models into embodied intelligent agents, they are evolving from task executors to autonomous decision-makers and expanding into complex collaborative tasks. However, the increasing intelligence and autonomy of these agents pose significant challenges to the robustness of swarm intelligence collaboration. While blockchain can provide trusted collaboration support, traditional energy-intensive architectures are constrained by the energy limitations of embodied intelligent agents, potentially leading to performance bottlenecks. To address this issue, this paper proposes DDSR, a consensus mechanism based on Dynamic Swarm Reputation, to provide trusted service support for embodied intelligence collaboration. DDSR confines the consensus scope within the collaborative swarm and links the agents’ stakes and fault tolerance thresholds to swarm reputation, ensuring trusted collaboration while achieving lightweight consensus. Additionally, DDSR introduces a dynamic fault tolerance strategy based on swarm reputation, allowing the fault tolerance threshold to be adaptively adjusted according to the reputation performance within the swarm, thereby optimizing consensus efficiency and security. Experimental results demonstrate that DDSR enables fair reputation evaluation, ensures the stability and security of the consensus process, and enhances the robustness of swarm collaboration.
With the rapid growth of multimedia content and online gaming, traditional cloud computing faces difficulties in meeting real-time requirements, particularly in optimizing caching strategies, managing dynamic environments, and handling user location uncertainties. To overcome these limitations, we propose a novel AI-Enhanced Edge Cooperation framework based on Multi-Agent Deep Reinforcement Learning (EC-MADRL) to optimize scheduling and resource allocation across edge nodes. This framework enables adaptive caching and replenishment strategies in a cooperative environment, modeled as a multi-agent Markov Decision Process (MDP). By integrating an online MADRL approach, the EC-MADRL algorithm allows edge nodes to collaboratively learn optimal policies for caching and resource distribution. We analyze the time complexity and convergence of the algorithm, demonstrating its effectiveness in improving edge node performance. Experimental results show a 30.12
Black-box attacks in deep reinforcement learning typically involve training substitute policies to imitate the behavior of target policies, crafting adversarial examples, and using these transferable adversarial examples to attack target policies. Previous works primarily study the transferability on non-targeted setting. However, recent studies show defects that lead to the difficulty in generating transferable targeted examples: noise curing. To address the above issues, we introduce a novel approach for targeted attacks that effectively generates more transferable adversarial examples. Our proposed method utilizes the Poincaré distance as a similarity metric, which allows for self-adaptive gradient magnitudes during iterative attacks and helps alleviate issues related to noise curing. Furthermore, we incorporate metric learning into the targeted attack process to steer adversarial examples away from the true action and enhance the transferability of targeted adversarial examples.
There are existing medical high-value consumables traceability solutions based on the SPD model. The centralized storage method has business pain points such as data insecurity, incomplete data, and lagging data reporting. By integrating multiple technologies such as blockchain + RFID + big data analysis, an application solution of blockchain and RFID technology in the traceability of medical high-value consumables is developed. This application not only improves the management efficiency of high - value consumables in hospitals, but also ensures the authenticity and security of data, completes the assetization of data, can better empower the refined operation of hospitals, and proposes a new example of “data element + intelligent management” application. It has been verified through experiments that what would take one working day to enter into the system using traditional manual methods can be completed in just five seconds with the introduction of RFID technology, significantly improving work efficiency.
The existing edge server deployment algorithms predominantly focus on the locations of base stations, often overlooking user experience. To address this limitation, this study proposes an edge server placement algorithm based on spectral clustering and Q-learning (QSC). The algorithm not only considers the locations of base stations but also incorporates the number of users and the geographical positions of base stations, with the aim of achieving a balanced workload distribution across edge servers while minimizing the average user access latency. The process begins with using spectral clustering (SC) to determine initial cluster centers, followed by applying the Q-learning algorithm to refine these centers, which are then designated as the deployment positions for the edge servers. Experimental results demonstrate that, compared to the traditional K-means algorithm, the QSC algorithm reduces access latency by 10.04
In recent years, the progress of deep learning has significantly promoted the development of RGB-D saliency object detection. However, existing methods are inefficient in feature processing. They either simply combine modalities through basic weighting or concatenation, failing to fully exploit complementary information, or they overlook local feature interactions, leading to detail loss. To address these issues, this paper proposes a more efficient approach during the encoder stage—the Cross-Modal Interaction Enhancement Module (CIE), which enhances local interactions to achieve more refined feature fusion, thereby effectively integrating RGB and depth features. During the decoder stage, we introduce the Hybrid Dilated Attention Fusion Module (HDA), which effectively processes three modal features—RGB, depth, and fusion—thereby enhancing the model’s performance. Experimental results demonstrate that our method outperforms other state-of-the-art approaches on four benchmark datasets: STERE, NJU2K, NLPR, and SSD, significantly enhancing performance.
Entity alignment, a critical process for integrating knowledge across diverse knowledge graphs, involves identifying and matching entities with identical semantics. The conventional approach relies on semi-supervised learning to align entities based on the similarity of their embeddings. However, the heterogeneity of data sources often results in non-isomorphic neighborhood structures for aligned entities, posing significant challenges, particularly for rare and sparsely connected entities. This paper introduces a novel soft label propagation framework that leverages multi-source data and iterative seed enhancement. The framework utilizes seed entities as anchors and selects optimal relational pairs to generate soft labels enriched with neighborhood features and semantic relationship information. Furthermore, considering the heterogeneity in neighborhood structures, a novel bidirectional weighted loss function is introduced to reduce the distance between positive pairs and to manage antagonistic pairs distinctively. This approach outperforms current semi-supervised methods, enhancing entity alignment accuracy across various datasets.
This paper addresses the challenges of verification efficiency in high-concurrency environments within digital asset trading platforms. We propose an optimized scheme based on Decentralized Identifiers (DID) and Schnorr aggregate signatures. By leveraging DID technology for decentralized identity management and employing an aggregate signature mechanism, our scheme significantly reduces the computational overhead of verifying multiple rights credentials, ensuring verification costs remain constant regardless of the number of credentials. Experimental results demonstrate that the proposed scheme offers substantial improvements in verification speed, achieving notable gains in large-scale transaction scenarios compared to traditional single-signature verification methods. Additionally, the scheme shows marked advantages in computational efficiency, storage overhead, and transaction throughput. The paper further explores the practical applications of the proposed scheme in digital asset trading, digital copyright protection, e-government, and supply chain finance. Our findings indicate that the scheme effectively enhances the efficiency and security of digital rights transactions, providing a promising solution for decentralized platforms.
With more and more interaction scenarios of cross-domain cooperation in different industries, the demand for information sharing and value exchange among organizations is increasing. How to obtain authentication from devices in other domains is one of the key technologies to realize data sharing among organizations. The traditional centralized authentication system is vulnerable to failure due to reliance on a single server. Existing solutions often overlook cross-domain authentication and suffer from inefficiencies caused by complex encryption processes, and they are not friendly enough for terminal devices with limited resources because many time-consuming cryptographic operations are involved. To address these issues, we propose a blockchain-based anonymous authentication scheme that realizes heterogeneous domain authentication with reduced computational overhead. We compare and analyze our scheme with the current researches in the same configuration environment. Experimental results show that the overall computational overhead of our scheme is reduced by 89
With the rapid expansion of cloud computing, the necessity for secure and efficient data sharing over the cloud has become more critical. Searchable Encryption (SE) technology plays a key role by enabling data owners to store encrypted data on the cloud while still permitting users to perform search operations. However, existing SE schemes encounter difficulties in maintaining a balance between ensuring data privacy, verifiability, and efficiency. To overcome these limitations, we propose OathSeek, a blockchain-based SE scheme that achieves all the above requirements. By maintaining version authentication tag, our approach significantly reduces the update overhead. Additionally, by leveraging blockchain technology, our scheme supports public verification of search results. We conducted a rigorous security analysis of the scheme and evaluated its efficiency through experiments. The experimental results demonstrate that our approach outperforms state-of-the-art (SOTA) schemes by approximately 3× .
With the rapid growth of large-scale datasets and complex AI models, federated learning has emerged as a promising approach to preserving data privacy. However, due to client heterogeneity and communication constraints, client selection serves as a crucial optimization problem. In real-world scenarios, varying data volume and dynamic client training states create a volatile environment. These factors lead to training delays and communication interruptions, significantly reducing the efficiency of federated learning. To address these challenges, a volatile-aware client selection method named CDE3 is proposed. First, we establish a multi-dimensional model to evaluate each client’s contribution. Then, the Exp3 algorithm is enhanced by incorporating a discount factor that exponentially weights historical contributions, and this refined algorithm assigns selection probabilities to clients based on their discounted historical contributions, enabling the server to make informed client selections. Finally, experimental results demonstrate that CDE3 effectively combats the volatility of the environment while maintaining high efficiency.
With the rapid development of technology, numerous service-oriented computing patterns have emerged, among which Blockchain as a Service (BaaS) has garnered widespread attention globally. As a crucial approach to blockchain application development, the proliferation of blockchain services has led to a large number of peers providing similar functionalities. Consequently, constructing a robust blockchain-based system necessitates carefully selecting dependable blockchain services (peers) that offer a superior quality of service (QoS). However, due to the vast array of available services and the limited availability of personalized QoS data, identifying the most suitable options poses significant challenges. Hence, we propose a novel graph neural network-based architecture, referred to as Double Attention (DBA). This framework integrates a multi-layer graph attention mechanism with self-attention to more effectively predict missing values in sparse scenarios. We conduct extensive experiments on a large-scale real-world dataset, and the results demonstrate that our DBA architecture not only achieves highly accurate predictions under sparse conditions but also outperforms existing approaches.
The security of electronic will notarization systems is increasingly being recognized by countries worldwide. Most existing schemes primarily rely on trusted third parties or blockchain-based time-lock puzzles (TLP) to ensure critical properties. However, these solutions often suffer from trust risks. To address these challenges, we propose a Secure Electronic Notarization Scheme with InterPlanetary File System (IPFS) and Blockchain (SENS-IFSB). Specifically, we introduce block-chain smart contracts as the core controller for document time-release encryption to mitigate security risks from third-party over-reliance. To enhance the flexibility of time-release mechanisms, we implement chained-Elliptic curve cryptography (ECC) encryption that enables adjustable extension of encryption time limits. Furthermore, we leverage the InterPlanetary File System to store time-lock puzzles and encrypted will contents, effectively alleviating blockchain storage pressure. Finally, we formally prove the security of IFSB in SENS-IFSB based on the hardness assumption of the Elliptic Curve Discrete Logarithm Problem (ECDLP). Our analysis demonstrates that the scheme is IND-CPA secure, with experimental results showing 50
The proliferation of dark web marketplaces has exacerbated illegal trade, particularly in narcotics, facilitated by Bitcoin’s pseudo-anonymous nature, which complicates transaction tracing and enables widespread illicit use. These platforms exploit the anonymity of the dark web and Bitcoin’s untraceability, posing significant regulatory challenges. Recent research has focused on analyzing Bitcoin-related criminal activities, including dark web transactions, Ponzi schemes, money laundering, and fraud. This study systematically investigates the operational mechanisms of dark web marketplaces and conducts targeted monitoring of a specific platform. Over a three-month period, Bitcoin transaction data was collected through active engagement and analyzed using machine learning-based detection methods. The proposed approach demonstrated high efficacy, achieving a true positive rate of 98.9
The metaverse refers to a virtual world that mirrors the real world, constructed based on technologies such as the Internet, extended reality, blockchain, and artificial intelligence. To enter the metaverse, users must have a digital avatar, i.e., a virtual identity. In metaverse services, identity authentication is a crucial element to ensure the trustworthiness of services. However, the virtual avatars or physical devices in the metaverse cannot determine the uniqueness of an entity’s identity. Moreover, the diversity of services in the metaverse makes it difficult to link identities across different scenarios, leading to the need for repeated authentication. Therefore, this paper proposes a distributed digital identity authentication scheme for the metaverse. We construct a dual-factor metaverse identity model based on Self-Sovereign Identity (SSI) and propose a scalable chain-based identity identifier using the Chameleon Signature Algorithm. Finally, we design a simulation experiment to test the efficiency and security of the scheme.
Artworks face copyright confirmation and protection challenges for their easy reproduction and dissemination, and traditional similarity detection methods are not robust enough under multiple image transformations, leading to difficulties in copyright confirmation and protection. In this paper, we propose an artwork copyright confirmation and protection method, DNACP, based on the DINOv2 model and NFT technology, which aims to solve the problems of attribution confirmation and piracy identification in traditional copyright and effectively avoid redundant NFTs. To ensure the originality of artworks, the DINOv2 self-supervised learning model is used to extract features from artwork images accurately, calculate the similarity between feature vectors, and then compare with a set threshold. A digital signature will be generated when the similarity does not exceed the threshold. At the same time, using NFT technology, the digital signature and copyright information of artworks are bound to the NFT, giving the work a unique digital identity and storing it on the blockchain, thus ensuring the non-tamperability and traceability of the copyright registration. The copyright confirmation experiment results prove the effectiveness of the DNACP method.
Air quality is closely related to modern environmental impact, carbon emission monitoring, and weather forecasting, playing a crucial role in maintaining ecological balance, addressing climate change, and safeguarding public health. High-quality air quality data is indispensable for accurate air quality prediction tasks, directly influencing urban planning, such as public facility placement and transportation planning. It also significantly affects residents’ travel decisions, such as outdoor sports choices and commuting methods. However, in actual monitoring, uncontrollable factors such as equipment failure, network interruption, and sensor contamination often lead to frequent missing data in air quality measurements. This severely threatens the accuracy of subsequent data analysis, the reliability of environmental monitoring, and the effectiveness of environmental governance. Therefore, air quality imputation has become a central concern in the field of air quality detection. Traditional statistical and machine learning-based imputation methods can address short-term data gaps, but they overlook the intrinsic temporal characteristics of air data, leading to large imputation biases. To address this, this paper proposes an air quality imputation model based on a multi-scale attention convolution network to achieve end-to-end imputation of air quality data. First, we introduce a novel multi-scale fractal method to transform one-dimensional air data into two-dimensional data samples with distinct periodic trends, providing a solid foundation for feature extraction. Second, we construct a carefully designed multi-scale hybrid convolution framework to deeply mine two-dimensional data features. Finally, through extensive experimental validation, we demonstrate that this deep learning-based imputation method effectively addresses air data gaps and achieves higher imputation accuracy than mainstream temporal imputation methods, providing strong technical support for air quality monitoring and management.
Recommender systems are pivotal in enhancing user experience on a variety of platforms by providing personalized suggestions based on user behaviors, historical interactions, and item attributes. Recently, recommendation methods combining graph neural networks with contrastive learning have gained attention for their superior performance. However, despite their success, they still face challenges in capturing structural information in the user-item interaction graph and are sensitive to noisy user behavior and the quality of contrast views. To address these limitations, this paper proposes a novel multi-view hypergraph contrastive learning method for item recommendation (VGCL). Specifically, the method involves constructing user-item interaction view, user-user collaborative view, and item-item collaborative view by leveraging user information, item information, and user-item interactions. It then learns user and item feature vectors by incorporating semantic and higher-order correlation information through graph and hypergraph neural network techniques. The method performs contrastive learning on the user and item feature vectors to enhance their quality. Finally, it predicts user preference scores for items using multi-layer perceptron and recommends the most suitable items to users. Extensive experiments on various real datasets show that VGCL outperforms existing state-of-the-art methods.
Test code is a tangible representation of software testing in practice and an important component of the project code. A good summary can help programmers quickly understand test code. However, we find that existing code summarization tasks often overlook the test code. They treat test codes as auto-generated codes and remove them from the dataset. Therefore, the performance of existing summarization model-dataset combinations on test code summarization is unknown. To obtain their performance and find the best model-dataset combination for test code summarization, we conducted an empirical study. The study was built on 5 deep-learning-based models, 3 datasets and 4 metrics. In particular, the 3 datasets include two widely used datasets that don’t contain test code, and the first JUnit Test code summarization dataSet we collected, JTS. The JTS contains 94,270 JUnit test-code-summary pairs extracted from 9,211 Java open source projects on GitHub. Since existing summarization models are already quite robust, while datasets almost all don’t include test code, we modified the datasets in our study. Specifically, in addition to exploring the performance of existing model-dataset combinations on test code summarization, we further investigated the impact of switching to datasets with reintroduced test code and specialized test code summarization datasets on performance. The experimental results show that existing model-dataset combinations are not well-suited for the test code summarization task. And the preferred model-dataset combination solution for this task is < pre-trained model, specialized test code summarization dataset > .
The market for Intelligent Connected vehicles (ICVs) is growing rapidly worldwide. One of the unique features provided by ICV is their companion mobile apps (called as ICV Apps) for end consumers, which facilitates functionalities such as remote vehicle control and management. However, previous research has shown that these apps generally lack transparency in their interactions with third-party services when collecting and sharing user data, raising significant privacy concerns among vehicle owners (app users). In particular, there are often inconsistencies between the privacy policies and third-party statements regarding the descriptions and intentions for the use of data items. In this paper, we propose PPChecker, a novel framework for automatically detecting inconsistencies in statements between mobile vehicle apps and their integrated third-party SDKs. More specifically, our research focuses on identifying two types of issues in privacy policy statements: missing data statements and ambiguous data statements. Both of these issues lead to privacy in-compliance of the mobile vehicle apps. To achieve this goal, PPChecker adopts an LLM-driven text analysis approach, specifically the Graph of Thought method, to identify potential inconsistencies. Evaluation results showed that PPChecker achieves 96.97
