Large Language Models (LLMs) have recently gained popularity for tasks like text generation and automated reasoning, often hosted on cloud servers to handle large-scale user interactions. Yet this setup faces a critical challenge, prompt injection attacks, where adversaries craft malicious text to bypass an LLM’s safety constraints and policy filters. Attackers constantly evolve new injection tactics, rendering static defenses ineffective. While service providers may wish to store user inputs for security audits, this practice poses serious privacy and data-protection risks. User prompts can contain personal or confidential information subject to strict regulatory requirements. To address these conflicting demands of security monitoring and user privacy, we propose a privacy-preserving prompt injection detection scheme via hash and embedding-based auditing for smart cloud-deployed LLMs. Our approach segments each user prompt into smaller chunks and stores only non-invertible representations of them—cryptographic hashes for exact matches and vector embeddings for near-match detection. This enables robust identification of both verbatim and paraphrased threats without retaining any plaintext data. Furthermore, newly discovered injection patterns can trigger retrospective audits on previously stored chunks, greatly enhancing attack traceability. We also introduce an optional blockchain-based logging mechanism for tamper-proof event records. Empirical results show that our method achieves high detection accuracy against diverse injection techniques while safeguarding sensitive user information.
With the growing integration of Smart Microgrids, cybersecurity concerns have become increasingly critical. Micro-grids face a variety of sophisticated cyber threats, including False Data Injection Attacks (FDIA), Denial of Service (DoS) Attacks, and Privilege Escalation Attacks, which compromise secure operations by obstructing communication networks and control systems. Existing security architectures lack the flexibility and adaptability to address the unique demands of microgrids. This paper proposes a novel Key Distribution Security Association-Based Confidentiality Isolation Control Framework, leveraging a dynamic GDOI multicast security association allocation method. The framework introduces fine-grained cryptographic boundary definitions and secure interconnection mechanisms to enhance microgrid security. By employing distinct Security Parameter Indexes (SPI) and integrating heartbeat signals, the framework achieves independent Security Associations (SA), effectively addressing issues like source authentication, replay attacks, and resource inefficiencies. Simulations and security analyses validate the proposed framework’s ability to defend against FDIA and replay attacks while reducing storage overhead and supporting dynamic adaptability in complex environments. This framework provides a scalable and practical solution for the secure operation of Smart Microgrids.
The rapid advancement of sixth-generation (6G) wireless networks has accelerated the adoption of terahertz (THz) communication due to its ultra-high data rates and abundant spectrum. However, distance estimation in non-line-of-sight (NLoS) THz environments remains challenging due to severe path loss, molecular absorption, and multipath effects. Moreover, precise localization raises privacy concerns, exposing nodes to adversarial tracking. To address these challenges, we propose a novel blind-distance estimation with adaptive privacy-preservation scheme for non-line-of-sight THz-based GPS-denied networks. We propose a novel blind-distance estimation method using a single THz frequency without requiring prior system knowledge. Second, we formulate a distance estimation model that integrates molecular absorption effects and multipath propagation to improve distance estimation accuracy. Third, we design a novel adaptive privacy-preserving perturbation mechanism that dynamically adjusts noise based on multipath propagation and obstacle density, ensuring feasible perturbations while balancing privacy and localization accuracy. Simulation results demonstrate that our scheme effectively balances localization accuracy and privacy preservation, adapting to both ideal and NLoS conditions in structured (urban) and unstructured (disaster) environments.
The integration of federated learning and recommendation systems is emerging as a prominent trend in machine learning, enabling personalized recommendations while preserving user privacy. In this paradigm, a master model is distributed to users, and the users perform local updates using their private data. The updates are sent back and aggregated on the server to update the master model then redistributed to the users. However, traditional federated recommendation systems encounter serveral challenges, including potential privacy leakage and high communication costs. To address these issues, we propose SIGN-FCF, a federated recommendation method based on matrix factorization, which leverages the SIGNSGD algorithm and differential privacy techniques. The proposed method employs a 1-bit compressor to enhance privacy protection and reduce communication costs, and three instances of the compressor are created to meet various privacy requirements. We also evaluate the performance of SIGN-FCF on three real-world datasets, demonstrating its effectiveness in preserving user privacy without compromising accuracy.
Many organisations utilise cloud-based data warehouses for their storage and analytics because these systems support multiple user tenants yet present considerable privacy threats. Traditional centralised data mining devices are sensitive to security risks, thus making federated learning (FL) an attractive solution for protecting information during mining processes. A research investigation analyses FedAvg and FedProx through a federated system while adding DP and secure aggregation technology to address security requirements. Tests demonstrate that SF-XGBoost delivers 99.31% accuracy alongside 0.9994 ROC-AUC values, which exceeds the other studied models in both centralised and federated learning platforms. The stability performance of FedProx is higher than that of FedAvg, although its precision and recall capabilities fall behind. Patronage clients that applied differentially private SF-XGBoost recorded 73.75% data accuracy while displaying privacy-utility configuration aspects. The slightly more significant losses of federated models parallel their ability to protect data privacy while satisfying privacy regulations. SF-XGBoost and privacy-improving methods deliver an effective security solution for cloud-based federated learning applications. Scientists must investigate how homomorphic encryption works alongside adaptive privacy budgets and federated transfer learning to achieve maximum optimisation of federated models.
In the current post-truth era characterized by an information explosion and complex and ever-changing circumstances, a large amount of false information and misleading content is widely disseminated, and the public's perception is easily disturbed. As a key mechanism for information screening and pushing, algorithms have seen their bias issues becoming increasingly prominent. In particular, the technological breakthrough of generative artificial intelligence has further transformed the infrastructure of the internet society. Its potential algorithmic bias and discrimination will not only solidify the systemic discrimination and inequality in human society, but may even intensify and create new biases, posing a systemic threat to social fairness and social progress. The algorithmic bias in the post-truth era has a non-negligible impact throughout the entire process of information production, information dissemination, and information construction. The ethical dilemma urgently needs to be addressed. This study reveals the generation logic and manifestations of algorithmic bias in the post-truth era, constructs a systematic path that integrates "technology, value, and institution" as a trinity, provides practical solutions to cope with the ethical challenges of algorithmic bias in the post-truth era, and helps build a more rational, inclusive, and just information society.
The explosive growth of distributed mobile devices has made federated learning (FL) a key privacy-preserving approach for collaborative training without sharing sensitive data. However, its untrusted decentralized nature exposes it to threats from malicious servers that can disrupt training or compromise the global model. Although secure aggregation methods are developed to mitigate these risks, many existing solutions rely on computationally heavy cryptographic techniques, introducing significant overhead. To overcome this limitation, we propose LiteSecFL, a lightweight and scalable secure aggregation framework for federated learning that tackles threats from malicious aggregators. The core of our approach is PolyAgg, a secure aggregation protocol that innovatively uses polynomial commitment with additive homomorphism to effectively defend against malicious aggregators. Furthermore, to reduce verification overhead, we incorporate a random sampling technique into our framework. Security analysis and experimental results across diverse models with different sizes, and various levels of sampling demonstrate the efficacy of LiteSecFL. The findings reveal that our LiteSecFL not only outperforms state-of-the-art methods by up to 9x in speed but is also scalable to different sizes of model architectures and the number of clients.
The Internet of Vehicles (IoV) is a core component of intelligent transportation systems in smart cloud, significantly enhancing traffic efficiency and reliability through real-time data sharing and collaborative decision-making. However, the high mobility of vehicles and frequent network topology changes complicate achieving secure and efficient data sharing in IoV. This work proposes an adaptive multi-party secure information-sharing framework based on digital twins (DT). Specifically, we introduce virtual representations of vehicles operating within an edge-centric DT architecture, thereby overcoming the limitations imposed by unstable physical links. Furthermore, we present a DT secure multiparty computation strategy based on a broadcast-request policy, which enables secure information sharing between vehicles in highly mobile and dynamic IoV scenarios. In addition, the proposed scheme incorporates blockchain to ensure the auditing of information sharing between virtual twin nodes. Simulation results using a consensus-based speed advisory system scenario validate the effectiveness of the proposed approach, demonstrating superior communication efficiency compared to conventional methods.
Cluster Computing Systems (CCS) is a type of technology that not only causes computing power improvement but also utilizes energy to a lesser degree by taking advantage of parallel programming while processing and reading massive amounts of data. We can have multiple Central Processing Units (CPUs) and storage devices (disks) where the massive size of data can be processed. However, Cluster Computing System also comes with its own set of challenges such as if for a reason the node stops operating, nodes stops communicating with each other and the data transfer doesn’t happen due to poor network which can lead to bottleneck while processing massive amounts of data. To overcome these issues, a well reputed tech giant known as Google, came up with a solution known as MapReduce. MapReduce is a framework designed for Big Data which takes care of processing large amounts of data over various servers. In this paper, we outline how CCS works and the challenges it faces today in the age of massive data. The introduction to some well received measures of Big Data are presented by us in this paper. These solutions show us the way we can address the issues we face in CSS. The primary goal of this writing is to look into the issues that we might face and the most efficient ways to resolve it in CSS.
IoT edge networks are constructed with Internet of Things (IoT) devices deployed near data sources at the network edge. These networks often include a middle tier of computationally robust devices designed to enhance IoT compute resources and boost network performance at an intermediary stage. By utilizing this middle tier for network capture, we introduce a novel side-channel attack through DNS pooling, AWS cloud services, and a rainbow table of AWS endpoint URLs. Our attack demonstrates the capability to correlate IoT edge devices with AWS cloud services, allowing us to determine a device’s role within the IoT edge network. This identification is crucial in reducing the time required to physically compromise the network by swapping out IoT edge devices with malicious counterparts. Furthermore, linking our attack to one or more AWS vulnerabilities amplifies the severity and scale of the exploit. Mitigating this technique is challenging as it exploits the inherent dependencies on DNS queries and the structure of AWS cloud services. Our experiment included a diverse set of IoT devices related to home automation and security. We accurately grouped IoT edge devices by analyzing traffic patterns, throughput, latency, cloud service endpoints, and human interaction patterns. Cloud service endpoint matching was achieved by utilizing AWS developer documentation, constructing EC2 pools through repeated DNS queries, and matching IP addresses to endpoints using a rainbow table of cloud service endpoints. Although AWS is highlighted in our experiment, other cloud services are similarly vulnerable due to their unique structural characteristics.
This paper presents a comprehensive analysis of secure data transmission technologies for low altitude unmanned aerial vehicles (UAVs). As UAVs become increasingly deployed in military, civilian, and commercial applications, the security vulnerabilities in UAV communication systems pose significant risks. We propose a multi-layered security framework that addresses the unique challenges of UAV data transmission, including lightweight cryptographic algorithms, physical layer security mechanisms, secure key distribution schemes, and anomaly detection systems. Our experimental results demonstrate that the proposed security solutions provide robust protection against common attacks while only increasing computational overhead by 5-8% and communication latency by 3-5%. Performance evaluations reveal significant improvements in resistance to man-in-the-middle attacks, replay attacks, and hijacking attempts compared to traditional security approaches. The framework's efficient design makes it particularly suitable for resource-constrained UAV environments while maintaining a high level of security assurance.
Natural language processing (NLP) is rapidly developing. A series of Large Language Models (LLMs) have emerged, represented by ChatGPT, which have made significant breakthroughs in natural language understanding and generation, enabling fluent dialogue with humans, understanding human intentions, and completing complex tasks. However, in addition to the fairness and toxicity of traditional language models, some new problems, including hallucination, have also emerged in LLMs, making them hard to use. Evaluating LLMs manually is challenging due to subjectivity and inefficiency. In this paper, we focused on the fuzzy matching, toxicity detection, and hallucination detection in the evaluation of LLMs automatically, and fine-tune the Mixtral-8x7B Model, which can be deployed in private cloud environment, and prove the effectiveness of our method through experiments.
Smart unmanned supermarkets have been emerging in the past years and significantly affected conventional shopping styles. Customer identification is an important step in picking up goods in smart unmanned supermarkets. However, in the actual application scenario, the camera can only capture part of the human body information, which is a challenge to identify individual consumers. This paper introduces an approach that integrates facial recognition with body re-identification (REID) algorithms to continuously update human feature information. The object detection method based on YOLOV4 is adopted to detect the human body in the unmanned supermarket. A full-scale network OSNet is used for human feature extraction. Finally, the cosine similarity of the extracted feature vector is calculated to obtain the similarity of human features. The experimental results show that the final similarity exceeds the set threshold value of 0.6, which can meet the needs of identity recognition for unmanned supermarkets.
Autism Spectrum Disorder (ASD) encompasses a range of complex neurodevelopmental conditions typically identified in early childhood. ASD is characterized by challenges in social interaction, communication, and by repetitive behaviors with restricted interests. The variability in symptoms' severity and expression among individuals presents significant diagnostic challenges to physicians. Advancements in computer technology have led various fields to adopt deep learning for constructing classification models. However, given the private nature of patient data, its leakage could have grave consequences. To mitigate this risk, we employ secure multiparty computing techniques and introduce a deep learning framework that ensures data interoperability without compromising privacy. Our framework facilitates deep learning training and inference via a lightweight, replicated secret-sharing technique. Experimentally, the scheme has been proven to exhibit high security, accuracy, and efficiency.
Cardiorespiratory synchronization training (CRST) uses breathing to increase balance in the autonomic nervous system and reduce negative emotions. Virtual Reality (VR) makes the patients immersive in the virtual environment for CRST training system. All the participants were trained to reduce their breathing rates through slow and deep breathing to achieve their target resonance frequency. Our proposed to design the new system based on virtual reality technology, and combined with a variety of neural sensors, such as Looxid EEG sensor, BMD101 ECG sensor, and Arduino SFM3400 Respirator, not only embedded on CRST training system, but also machine learning method to stimulate a real relaxing environment. Develop the innovative features of clinical psychological intervention, user convenience and provide the therapist with real-time display of heart rate features(SDNN, LF, LnLf, HF, LF/HF, Mhr), detect emotion and flow data on the monitor. During the process, the computer calculates LnLf value and change VR scenes at any time, we provide an instant interactive system. Our showed the two modules of CRST training system including both of passive CRST module and direct CRST module, that heart rate variability (HRV) variables and EEG have significant difference between the first one minute and last one minute in CRST VR training system.
With the rapid development of industrial production, defect detection has become increasingly important across various fields. However, traditional defect detection methods suffer from issues such as low efficiency, high costs, and lack of real-time capabilities. To address these problems, this paper proposes an intelligent defect detection system based on cloud-edge synergy. This system leverages the advantages of cloud computing and edge computing to achieve efficient, cost-effective, and real-time defect detection. Specifically, the system first trains models on cloud servers and then distributes the trained models to edge nodes for real-time defect detection. By utilizing cloud-edge synergy, the system can fully optimize the computing power of edge computing and the storage capacity of cloud computing, thereby enhancing the overall performance. Through experiments, this paper demonstrates the superior performance of the system in different domains through experiments and provides a detailed performance evaluation.
The sequence of events (SOE) recording is an important information for analyzing power system faults. If the clock system in the substation is not synchronized properly, it can lead to errors in SOE recording time, affecting the analysis and judgment of faults. On the basis of analyzing the implementation method of clock synchronization in the scheduling automation system, this article proposes a method based on fuzzy neural network for detecting synchronization anomalies in the substation clock system. This method can detect anomalies in the clock synchronization system in the substation at the first time. The experiment shows that this method can greatly improve the timeliness of synchronous fault detection and fault handling efficiency in the substation clock system, further enhancing the reliability of the clock synchronization system and substation automation system.
As an emerging paradigm of content creation, AI-Generated Content (AIGC) has been widely adopted by a large number of edge end users. However, the requests for generated content from AIGC users have obvious diversity, and there remains a notable lack of research addressing the variance in user demands for AIGC services. This gap underscores a critical need for suitable AIGC service selection mechanisms satisfying various AIGC user requirements under resource-constrained edge environments. To address this challenge, this paper proposes a novel Attention-based Diffusion Soft Actor-Critic (ADSAC) algorithm to select the appropriate AIGC model in response to heterogeneous AIGC user requests. Specifically, the ADSAC algorithm integrates a diffusion model as the policy network in the off-policy reinforcement learning (RL) framework, to capture the intricate relationships between the characteristics of AIGC tasks and the integrated edge network states. Furthermore, an attention mechanism is utilized to harness the contextual long-range dependencies present in state feature vectors, enhancing the decision-making process. Extensive experiments validate the effectiveness of our algorithm in enhancing the overall user utility and reducing the crash rate of servers. Compared to the existing methods, the proposed ADSAC algorithm outperforms existing methods, reducing the overall user utility loss and the server crash rate by at least 58.3% and 58.4%, respectively. These results demonstrate our ADSAC algorithm is a robust solution to the challenges of diverse and dynamic user requirements in edge-based AIGC application environments.
As intelligent transportation systems continue to develop, the integration of cloud-assisted device-to-device (D2D) communication and trajectory prediction has become a crucial strategy for enhancing traffic management and safety. However, D2D communication faces challenges such as bandwidth limitations and high latency, which are particularly pronounced in dense urban environments. This paper introduces a novel framework that utilizes self-attention mechanisms and edge content caching to optimize D2D communication and trajectory predictions. By leveraging cloud resources to alleviate computational burdens and employing edge caching to reduce latency and bandwidth consumption, our approach ensures that data transmission between vehicles is both fast and economical. Our innovative method facilitates real-time, accurate trajectory forecasting and efficient data communication among vehicles. We have developed a self-attention model that dynamically prioritizes relevant data points and trajectory information based on historical and contextual traffic data, thereby enabling more precise predictions and a robust communication network. To validate the effectiveness of our proposed framework, we compared its performance with traditional methods. The results demonstrate significant improvements in predictive accuracy and communication efficiency compared to conventional approaches.
Federated Learning (FL) protects data privacy by sharing gradients across clients rather than local training data. It has always been a hot research issue to motivate users to actively contribute local data and participate in the federated learning aggregation process. This paper proposes a novel Mean-Field-Game-based Federated Learning incentive mechanism. We first model the process of federated learning aggregation as a mean-field game problem across clients. We then design a mean-field federated learning gradient calculation algorithm based on stochastic differential equations, i.e., HJB and FPK equations. We build an efficient client reputation-aware incentive mechanism that improves global learning performance by comparing the cosine similarity of the obtained mean-field and individual FL gradients. Finally, experimental results show that our incentive mechanism outperforms the baseline algorithms in FL learning performance.
