Microservice-based edge computing decomposes applications into lightweight services that can be deployed near users to reduce service latency. However, practical edge environments must balance latency reduction and edge-serving efficiency: aggressively minimizing latency may overload limited edge resources, whereas prioritizing edge utilization may increase congestion and degrade service quality. This paper studies the multi-objective microservice deployment problem (MOMDP), which jointly minimizes the average waiting time of service requests and maximizes the fraction of requests served by edge servers. We formulate MOMDP as a mixed-integer linear programming (MILP) problem under demand conservation, service-placement coupling, coverage, CPU, memory, and cloud offloading constraints. Since solving the MILP online is impractical under dynamic workloads, we propose an optimization guided deep deterministic policy gradient (DDPG) framework. The framework uses a relaxed optimization layer to project actor generated deployment and resource-allocation preferences into feasible executable decisions, enabling adaptation to time-varying demand while satisfying system constraints. Extensive simulations across multiple demand profiles show that DDPG+Optimization achieves lower average waiting time, higher edge-serving ratios, and reduced cloud dependency compared with the baseline methods.
The increasing deployment of edge-level devices in industrial control systems is increasing the interaction with public networks and intelligent electronic devices. This increasing deployment is introducing new and complicated threats to autonomous energy systems (AES) and other interdependent critical infrastructure. Combined data falsification attacks (CDFAs) are a critical category of cyberattacks that can adversely affect the operation of AES. CDFAs are more complicated than traditional false data injection attacks because attackers can simultaneously manipulate and blind measurements on supervisory control and data acquisition systems to mislead control operations while remaining stealthy. Detecting CDFAs in AES is a challenging task caused by the complexity of data properties and low detection accuracy. This work proposes a novel method for detecting CDFAs executed in AES considering the correlation and redundant characteristics of the compromised data in the AES. An unsupervised attention-based network, a long short-term memory (LSTM) autoencoder, is proposed to detect CDFAs considering the nonfunctional dependency and spatiotemporal correlation of AES operational data. The efficacy of the proposed algorithm is evaluated and verified on a standard IEEE 39-bus system and a 2,000-bus synthetic grid on the footprint of Texas using synthetic phasor measurement unit data. The results show that the proposed method can effectively improve the detection accuracy and effectiveness of CDFAs on AES. The performance of the proposed method is also compared with a one-class support vector machine. Finally, we explain the ability of the LSTM autoencoder model to detect events by using SHapley Additive exPlanations (SHAP) as post hoc explanations. This explanation interprets why and how the model is using specific aspects to make predictions, validating the actionable adoption of this detection technique.
Blockchain technology is increasingly adopted in data-intensive industries. As blockchain data grows rapidly, users with limited storage and computational resources often participate as light nodes, relying on third-party service providers to store the full blockchain data and offer query services. In such scenarios, where service providers may maliciously tamper with data, ensuring the integrity of query results becomes critical. Additionally, graphs are widely used to model relationships among real-world entities. Identifying cohesive subgraphs, such as core subgraphs, is a fundamental problem in graph analysis, and applications in blockchain include anti-money laundering and account-based sharding. However, to the best of our knowledge, no existing research addresses verifiable core subgraph queries in blockchain environments. Key challenges include representing graph structures, validating subgraphs, and handling diverse graph types (such as one-dimensional, multi-dimensional, undirected, and directed graphs). In this paper, we introduce the concept of the partial neighbor set to capture graph structures, which eliminates edge redundancy and enables efficient search for associated edges. Building on this, we propose novel authenticated data structures (ADSs) that support both one-dimensional k-core and multi-dimensional m-core queries in directed and undirected graphs. These ADSs utilize vertex coreness as a search key and integrate cryptographic set accumulators and Bloom filters to efficiently generate proofs of mismatched vertices and edges in the graph. We further improve search efficiency and proof aggregation by introducing batch search and verification. Experimental results demonstrate the efficiency and effectiveness of the proposed approaches, especially for large graph data.
Federated learning, as a new technology to protect data privacy, effectively solves the problem of data silos. To solve the problem of data heterogeneity in distributed architecture, this paper proposes a personalized federated learning method called FedPL. It is based on iterative partitioning and parameter hierarchy. Firstly, to address the slow model convergence caused by non-independent and identically distributed (Non-IID) data, and considering the advantages of clustering in resolving weight divergence, this paper proposes an iterative partitioning method based on a ternary cosine similarity measure. In the process of model training, the proposed method dynamically adjusts the division results according to the similarity of parameter updates. It realizes the rapid convergence of the local model and thus guides the rapid convergence of the global model. Secondly, aiming at the high complexity of federated computing, considering the advantage of SVD in reducing the computational complexity of massive data, a method of computing similarity based on SVD is proposed. This method performs dimensionality reduction on the base level parameters of the clients, achieving improved efficiency in similarity computation. Finally, to address the problem of local distributional differences, this paper proposes a personalized federated learning method based on parameter layering, taking into account that different preferences can be captured from clients' raw data. This method trains the base layer on the global data and the personalized layer on the local data respectively. It helps mitigate the issues arising from local heterogeneity, thereby achieving personalized customization for client models. The experiments demonstrate that this method can accelerate the federated training process while providing personalization for individual clients, thereby enhancing the adaptability of local models.
Internet of Things (IoT) technology has revolutionised data management. Cloud-assisted IoT solves the problem of limited storage and computing power of IoT devices, but data security issues, such as confidentiality, and authenticity remain a huge challenge. Multi-authority attribute-based signcryption (MA-ABSC) can effectively solve this problem and avoid the single-point bottleneck problem in attribute-based signcryption (ABSC) with a single authority. However, in existing MA-ABSC schemes the length of the user secret key and the ciphertext increases linearly with the number of authorities, which leads to more storage requirements for IoT devices. This paper proposes an efficient MA-ABSC scheme, in which both the length of the user secret key and the ciphertext are independent of the number of authorities. We also outsource part of the unsigncryption work to the edge servers to reduce the computation cost. In addition, our scheme enables efficient revocation of data users. Security analysis proves that MA-EABSC is indistinguishable under adaptive chosen ciphertext attacks, and existentially unforgeable under adaptive chosen message attacks. The experiments show that our scheme is very efficient and practical for edge server assisted IoT.
Point-cloud-based CAD reverse engineering aims to recover editable parametric models from scanned geometry. Existing methods remain limited by inaccurate effective parameter prediction, redundant point-cloud inputs, and costly task-specific training. To address these limitations, we propose LiCoCAD, a lightweight model–LLM collaborative framework that reconstructs point clouds as CADQuery code. It uses lightweight modules to generate key constraints and prompts for LLM-based CAD reconstruction. The lightweight CAD-Coder predicts command sequences as structural constraints, while a training-free geometry keypoint extractor converts scanned point clouds into compact coordinate prompts as geometric constraints. These structural and geometric constraints guide commercial LLMs to reason about CAD parameters and generate CADQuery code. This design avoids task-specific LLM fine-tuning and trains only the lightweight CAD-Coder, thereby shifting the main training burden to lightweight task-specific model training. During inference, a CD-feedback refinement stage enforces code validity and refines geometric parameters. Experiments on three datasets show that LiCoCAD achieves higher reconstruction accuracy and robustness than existing methods. Real-world industrial cases further show that the generated CADQuery code remains editable. These results suggest that the lightweight model–LLM collaborative paradigm provides a scalable solution for CAD reverse engineering and is transferable to broader engineering tasks.
Copyright protection in Federated Learning (FL) environments faces some challenges, such as fidelity, validity, robustness, and persistence, due to the large number of diverse FL clients. To address these issues, we propose a Grouped Clients Hierarchical Watermarking (GCHW) scheme, which consists of client grouping, watermark generation, watermark embedding and watermark verification. The client grouping is to group clients based on shared characteristics. In watermark generation, a hierarchical framework is proposed to establish dependencies between the group and client watermarks by embedding part of the group watermark information into each client water mark. Moreover, our GCHW scheme innovatively embeds binary watermarks into the frequency domain of model parameters at both the group and client levels during the watermark embedding phase. Experimental results demonstrate that GCHW effectively maintains fidelity and the validity of the watermarks is validated across independent and identically distributed (IID) and non-IID scenarios. GCHW also exhibits strong robustness against common attacks and defensive aggregation algorithms. In addition, GCHW shows good persistence in dynamic FL environments.
Energy theft, characterized by manipulating energy consumption readings to reduce payments, poses a dual threat—causing financial losses for grid operators and undermining the performance of smart grids. Effective Energy Theft Detection (ETD) methods become crucial in mitigating these risks by identifying such fraudulent activities in their early stages. However, the majority of current ETD methods rely on supervised learning, which is hindered by the difficulty of labelling data and the risk of overfitting known attacks. Several unsupervised ETD methods have been proposed to address these challenges, focusing on learning the normal patterns from honest users, specifically the reconstruction of input. However, our investigation reveals a limitation in current unsupervised ETD methods, as they can only detect anomalous behaviours in users exhibiting regular patterns. Users with high-variance behaviours pose a challenge to these methods. In response, this paper introduces a Denoising Diffusion Probabilistic Model (DDPM)-based ETD approach. This innovative approach demonstrates impressive ETD performance on high-variance smart grid data by incorporating additional attributes correlated with energy consumption. The proposed methods improve the average ETD performance on high-variance smart grid data from below 0.5 to over 0.9 w.r.t. AUC.
The unprecedented growth of computer vision applications, such as surveillance systems and social media, raises security and visual privacy concerns, especially when data is stored on cloud servers. Image obfuscation offers a way to preserve visual privacy while maintaining an adequate level of usability; thus, it has been a topic of great interest in recent years. However, prior obfuscation schemes are either vulnerable to malicious attacks, such as model inversion to reconstruct original images from obfuscated images, or generate non-trainable obfuscated images, making them unusable for achieving reasonable accuracy. This paper proposes a novel bit-plane-based image obfuscation scheme, Bit-ViP, to preserve visual privacy for image-based recognition tasks. The Bit-ViP scheme produces secure, usable images by incorporating an innovative end-to-end obfuscation function. While doing so, the obfuscated image would contain non-invertible noise (generated by Lorenz's chaotic system and differential privacy), making it hard for an adversary to reconstruct the original image. We conduct extensive experiments on two popular activity recognition datasets, namely UCF101 and HMDB51, to validate the effectiveness of Bit-ViP. In the face of attacks on reconstruction, pixel frequency, information entropy, and pixel inter-correlation, we present a rigorous security analysis demonstrating tangible improvements over existing schemes.
Increasing application memory demands have led to the emergence of tiered memory systems, which place significant pressure on local memory nodes. The fundamentals of such systems involve selecting promotion candidates based on page access behaviors (e.g., frequency and recency) and migrating these candidates between memory tiers to leverage both low access latency and large capacity. In this paper, we examine the practical challenges of tiered memory systems, particularly under severe memory pressure conditions. Specifically, this study makes the following key observation: Tiered memory systems can suffer slowdowns from existing on-demand synchronous promotion methods, as they continue to perform promotion candidate selection processes even when local memory is unavailable. Furthermore, such excessive candidates have an equal chance of being promoted, and relatively more frequently accessed pages lose opportunities for successful promotion. To address these issues, we propose SAPPhire, a system that prioritizes local memory availability when selecting promotion candidates and conditionally ensures the successful promotion of hot pages. The complementary application of these two policies effectively enhances performance in both kernel and user spaces. As a result, experimental results demonstrate that SAPPhire outperforms six prior page migration methods by an average of 1.2×.
Resistive Random Access Memory (RRAM) has emerged as a promising candidate to succeed conventional memory technologies, offering advantages such as low energy operation and scalable storage density. As different manufacturers continue to advance RRAM toward large-scale production and commercial deployment, the demand for accurate and efficient testing methodologies becomes increasingly critical. This paper investigates interconnect and contact defects in RRAM arrays, with particular attention to the influence of the Data Background (DB), and proposes corresponding test strategies. A comprehensive defect space covering interconnect and contact defects is formulated for a layout-independent RRAM design. Systematic defect injection and circuit-level simulations are conducted to derive fault models, encompassing not only single-cell and two-cell coupling faults but also multi-cell faults strongly affected by DBs. The analysis identifies previously unreported 3-cell and 4-cell faults, which originate from defect-related sneak paths in the array. Conventional RRAM test schemes are unable to detect such faults. To address this limitation, a DB-aware test generation scheme is presented, capable of efficiently identifying all the aforementioned fault types and thereby enhancing the overall fault and defect coverage in RRAM devices.
Maximal Extractable Value (MEV) poses significant threats to the security and fairness of Ethereum's decentralized finance ecosystem by enabling participants to exploit transaction ordering to extract profits at the expense of others. Heuristic-based detection methods have been widely adopted to identify MEV transactions such as sandwich, arbitrage, and liquidation. However, a lack of standardized evaluation across heuristics limits the ability to compare their detection behavior and computational characteristics. This paper presents a unified experimental framework to systematically evaluate the detection capabilities, agreement rates, and resource efficiency of existing heuristic approaches. An agreement metric is introduced to quantify consistency across detection methods. Additionally, we assessed resource utilization and execution time to evaluate computational scalability. Our empirical findings reveal that while these approaches exhibit scalability, their reliability, measured in terms of detection consistency, varies significantly across different MEV types. Agreement rates averaged 0.53 for sandwich detection, 0.40 for arbitrage, and 0.81 for liquidation, highlighting substantial differences in detection capabilities and heuristic formulations. These findings offer valuable insights into the practical challenges of achieving consistent MEV detection and highlight the need for developing more robust security countermeasures.
This paper presents sGprAle, a self-supervised framework for predicting Average Localisation Error (ALE) in wireless sensor networks under limited labeled data conditions. The proposed approach integrates Gaussian Process Regression (GPR) with an autoencoder-based feature learning strategy to enhance robustness, generalisation, and predictive reliability. By learning task-agnostic latent representations from unlabeled data through a reconstruction objective, sGprAle enables the GPR model to operate in a compact and informative latent space using only a small labeled subset. Bayesian optimisation is employed to adapt kernel hyperparameters, ensuring calibrated uncertainty estimates and stable performance when labeled data are scarce. The framework is systematically evaluated across multiple labeled-data scenarios using ablation studies, uncertainty analysis, and statistical comparisons. Results demonstrate that the self-supervised latent representation substantially improves model robustness and label efficiency, allowing sGprAle to achieve competitive accuracy relative to existing methods while requiring significantly fewer labeled samples. These findings highlight the effectiveness of combining self-supervised representation learning with probabilistic regression for addressing the node localisation problem. The proposed model offers a reliable and energy efficient solution for ALE prediction, contributing to improved localisation performance and operational efficiency in wireless sensor networks.
Distributed consensus (DC) protocols typically assume a deterministic fault tolerance model, such as requiring fewer than one third of the nodes to be faulty in Byzantine Fault Tolerance (BFT). However, this assumption is often impractical in real-world scenarios, where node reliability is probabilistic, making fixed thresholds rigid and inflexible. Probabilistic Consensus (PC) offers an alternative by replacing these fixed limits with probability-based non-Byzantine behavior models. Yet, existing studies primarily focus on system performance while overlooking the implications of this assumption on DC mechanisms. In this paper, we address two critical issues. First, we systematically analyze the non-deterministic outcomes of PC caused by probabilistic node reliability, focusing on two key dimensions: consistency, which ensures agreement among nodes, and correctness, which ensures that consensus decisions are unaffected by faulty nodes. We propose both a rapid estimation method and a detailed probabilistic assessment. Second, we redesign quorum selection in PC by introducing the reliability quorum, the minimal subset of nodes required to achieve a specified system reliability level. Finally, we validate our theoretical findings through simulations.
Today's data-centric applications demand cache architectures that can scale with growing workloads while maintaining high performance and energy efficiency. Fundamental issues such as excessive area consumption and leakage power are increasingly challenging traditional SRAM-based caches, thereby motivating the exploration of non-volatile alternatives. Among these, racetrack memory (RTM) stands out due to its remarkable storage density, achieved through nanowires hosting sequential magnetic domains that can be manipulated via domain wall or skyrmion techniques. Despite its advantages, racetrack memory's inherent serialized access introduces considerable shift overhead, leading to elevated energy consumption and latency. In this paper, we analyze the placement strategies in conventional RTM-based last-level caches and identify that current methods trigger redundant shift operations as a shift intended to read one block fails to preposition other blocks subsequently accessed. To resolve this, we introduce an innovative data placement and replacement scheme that intelligently groups correlated blocks, ensuring that a single shift not only retrieves the target block but also aligns subsequent blocks closer to the access port. Our simulation results using the gem5 simulator and the SPEC CPU2017 benchmarks reveal that our scheme reduces shift overhead by 49.0
Unintentional or intentionally faults into hardware accelerators of Post-Quantum Cryptography (PQC) or Fully Homomorphic Encryption (FHE) algorithms may lead to the leakage of sensitive information. If appropriate fault protection mechanisms are not employed, recent PQC schemes such as Kyber, Dilithium, Falcon, and NTRU, as well as FHE schemes such as Brakerski-Fan-Vercauteren (BFV) and Brakerski Gentry Vaikuntanathan (BGV), may leak sensitive information at the hardware implementation level. This work proposes three efficient recomputationbased fault detection methods for Barrett Modular Multiplication (BMM) in the Cooley-Tukey Butterfly Unit (CT-BU) of Number Theoretic Transform (NTT ), implemented on a Field-Programmable Gate Array (FPGA) for PQC and FHE algorithms. The CT-BU and BMM are fundamental components in lattice-based PQC and FHE schemes. This paper introduces a new algorithm, Recomputation with Swapped Operand (RESWO), for fault detection. While Re-computation with Negated Operand (RENO) and Recomputation with Shifted Operand (RESO) are existing methods used in other PQC hardware algorithms. To the best of our knowledge, RENO and RESO have never been used in wordwise BMM before. The proposed RESWO method consumes a comparable number of slices to RENO and RESO. However, for most PQC and FHE variants, RESWO achieves lower energy consumption and reduced slice utilization compared to both RENO and RESO. The fault detection efficiency of RESWO, RENO, and RESO is nearly similar to 100%.
Healthcare federated learning faces institutional diversity, data constraints, and computing resource discrepancies among other difficulties. These problems call for creative solutions for fair and effective distributed medical image analysis that improves model performance while handling the unique data features in different healthcare environments. This paper presents RAGE-FED (Resource Aware GEneration for FedErated Disease Diagnosis), a new paradigm meant to address these important problems. Emphasising computational efficiency, communication overhead reduction, and energy saving, RAGE FED seeks to maximize performance across healthcare facilities while ensuring diagnostic accuracy and privacy. To achieve these objectives the framework makes use of advanced data generation methods and adaptive training systems. To effectively use the different processing capacity of the participating nodes, it uses dynamic model complexity adjustment and adaptive learning rate scheduling. To protect private patient data, RAGE-FED additionally uses privacy-preserving techniques like differential privacy and safe aggregation. By use of synthetic data creation, the framework also solves data shortage and imbalance problems. The performance evaluation on the Chest X-Ray Pneumonia dataset confirms the effectiveness of RAGE-FED in diverse healthcare settings, showing gains of over 21.47% in efficiency, communication, and energy use compared to state-of-the-art methods. Extended experiments on additional datasets further validate generalizability, with improvements of 20.13% on Brain Tumor MRI, and 17.85% on HAM10000 skin lesion images. Together with ablation and sensitivity analyses, these results demonstrate the robustness, scalability, and broad applicability of the proposed framework.
