This paper presents a scalable computingcontinuum framework for Ambient Assisted Living (AAL)-based IoT applications. It incorporates diverse Internet of Things (IoT) data sources, including environmental sensors, facial-recognitionenabled indoor drones, depth cameras, and monitoring robots. These devices operate across a continuum comprising edge devices, an edge server, and a cloud layer. The framework is evaluated through a Large Language Model (LLM)-based web application that accepts natural-language queries and retrieves data from IoT sources connected to the continuum. Experimental results demonstrate that latency-sensitive and privacy-critical data is stored and processed at the edge, while computation-intensive or non-sensitive data are offloaded to cloud. This Quality of Service (QoS)-aware data placement and query-routing strategy enables scalable operation across heterogeneous compute units. This approach sustains throughput under concurrent workloads while delivering near real-time responses. Evaluation results for data retrieval and delivery for up to 150 simultaneous queries show end-to-end response times of under 3.5 seconds at the edge layer and under 10 seconds at the cloud layer, demonstrating the framework's scalability.

Low Earth Orbit (LEO) satellite constellations provide a wide range of services such as communications, earth observation, signal monitoring, and scientific missions. While these constellations generate valuable data, transferring it to ground stations (GS) for machine learning-based analysis presents significant challenges due to downlink bandwidth and energy constraints. Federated Learning (FL) integrated with Orbital Edge Computing (OEC) has been explored as a solution to these challenges. This paper presents a novel energyefficient and decentralised FL framework designed to optimise communication with GSs and minimise energy consumption. The performance evaluation conducted under the Walker Deltabased LEO constellation across various datasets reveals that our approach can sustain high accuracy while considerably reducing training time and energy consumption.

Low-Earth-orbit (LEO) satellite systems are evolving into distributed service platforms with onboard intelligence, creating demand for on-orbit training services in Earth observation, disaster response, and autonomous space applications. However, weakly labeled or unlabeled sensing data, intermittent ground-station (GS) visibility, heterogeneous onboard resources, and strict timeliness requirements make collaborative on-orbit training a QoS-aware service orchestration problem rather than only a distributed learning problem. To address this challenge, we propose OrbitServe, a QoS-aware orchestration framework for label-scarce on-orbit training across LEO satellite systems. OrbitServe adopts a two-stage design. In the planning stage, it provisions executable training services through lightweight participant profiling, master provisioning, and capacity-aware clustering. In the runtime stage, it enables master-led semisupervised collaboration and hierarchical synchronization under intermittent connectivity, reducing repeated GS coordination while supporting practical on-orbit execution. By jointly coordinating service provisioning and runtime execution, OrbitServe improves the trade-offs among model accuracy, job completion time (JCT), and energy consumption. Experiments on satellite image datasets show that OrbitServe consistently outperforms representative baselines, achieving up to 95% accuracy, reducing JCT from hundreds of days to within 24 hours, and lowering total energy consumption by about one order of magnitude at comparable accuracies.

Graph-structured data are integral to applications like social networks, biological systems, cybersecurity, and fraud detection. Outsourcing these data to public clouds offers scalability but raises privacy concerns, as encryption is required before outsourcing, making traditional graph similarity search and access control challenging. This paper presents a novel solution for privacy-preserving full graph similarity search with fine-grained access control in cloud environments. To the best of our knowledge, this is the first work to integrate privacy-preserving graph similarity search in a multi-user/multi-query setting with attribute-based access control (ABAC). This enables scalable and secure access in realistic, collaborative environments. The graph owner leverages the neural Graph2vec model to create feature indexes for encrypted graph data. Simultaneously, a secure transfer learning mechanism enables graph users to generate query feature indexes in the same latent space, ensuring privacy while accurately capturing the user’s query intent. ABAC is employed to enforce flexible, fine-grained access policies. We conduct a formal security analysis under known-ciphertext and known-background threat models, demonstrating strong privacy guarantees. Experimental evaluations on real-world datasets show that our scheme achieves high semantic accuracy, lower search latency, and reduced storage overhead, outperforming existing approaches.

The proliferation of distributed energy resources and multi-energy coupling technologies has transformed virtual power plants into complex MEVPP clusters. However, coordinating these heterogeneous clusters poses multifaceted challenges, particularly due to the deep stochastic uncertainties of renewables, the complex coupling of energy vectors with distinct dynamic inertias, and the severe computational bottlenecks inherent in traditional centralized control across multiple timescales. This paper proposes a novel hierarchical control strategy that leverages cloud-edge collaborative architecture and digital twin technology to optimize MEVPP clusters across electricity, heat, and gas networks. The framework employs a three-layer temporal decomposition: cloud-based day-ahead scheduling using mixed-integer linear programming for 24-h economic dispatch; edge-based intra-day rolling optimization with 15-min intervals for uncertainty mitigation; and real-time DMPC with 5-min resolution for dynamic balancing. A comprehensive digital twin framework integrates physics-based multi-energy flow models with data-driven techniques to enhance state estimation and prediction accuracy while maintaining computational efficiency for edge deployment. The DMPC algorithm coordinates multiple MEVPPs via decomposition and coordination, managing cross-coupling constraints and optimizing power distribution via tie-line sharing. Case studies on three heterogeneous MEVPPs demonstrate that the proposed strategy achieves a 1.07% reduction in total system costs through coordinated operation compared to independent optimization, with individual MEVPP cost reductions ranging from 8.77% to 28.84% during intra-day operation under renewable forecast uncertainties, while maintaining system stability and inter-MEVPP power exchange balance.

Low Earth Orbit (LEO) mega-constellations extend the cloud-to-edge continuum into space, enabling satellite edge computing. However, Federated Learning (FL) in this environment is fundamentally energy-constrained due to dynamic inter-satellite connectivity, heterogeneous onboard computing hardware, and strict power budgets. We propose CroSatFL, a sustainable on-orbit hierarchical FL framework that reduces end-to-end energy across computation and communication while maintaining strong training performance under realistic LEO dynamics. CroSatFL keeps the ground station (GS) off the iterative loop by performing all local training and intermediate aggregations on orbit, requiring only two GS communication phases: one for initialization and one for final model collection. This sharply reduces repeated use of bandwidth-limited and energy-expensive GS links and shifts iterative exchanges to laser inter-satellite links (LISLs). CroSatFL integrates three energy-aware mechanisms: StarMask forms LISL-feasible clusters that align data volume with heterogeneous CPU/GPU capability, Skip-One mitigates transient stragglers by skipping at most one slow client per cluster to lower round energy and latency while preserving long-term fairness, and random-k cross-aggregation enables lightweight topology-aware cross-cluster mixing without extending round duration. Using an end-to-end energy model with a realistic Walker-Delta constellation, we show that CroSatFL reduces GS communication count by over two orders of magnitude and GS transmission energy by about 6x relative to GS-centric and on-orbit baselines, while achieving competitive accuracy and faster convergence.

Federated learning (FL) enables collaborative model training across distributed clients while preserving data privacy, but non-independent and identically distributed (non-IID) client data can amplify algorithmic bias and raise significant fairness concerns. This paper presents a systematic empirical evaluation of clustered federated learning (CFL) strategies, analyzing accuracy and fairness across different numbers of clusters $(k)$ and controlled levels of heterogeneity generated via Dirichlet partitioning with concentration parameter $\alpha$. By varying these parameters, we study behavior from extreme class imbalance to near-IID client distributions under different demographic contexts. We compare four representative strategies: FedAvg, IFCA, FL+HC, and WeCFL, on the COMPAS (race-based fairness) dataset and validate key patterns on the Adult (gender-based fairness) dataset, using predictive accuracy and five complementary fairness metrics. Overall, clustering can improve both fairness and accuracy in highly heterogeneous regimes, but benefits diminish as data become near-IID and some methods (notably FL+HC) exhibit instability under specific configurations; one-way ANOVA and Tukey HSD confirm that strategy choice significantly affects accuracy, while increasing clusters beyond $k=5$ yields limited additional gains, providing practical guidance for deploying heterogeneity-aware FL systems that seek measurable fairness improvements without sacrificing performance.

The exponential growth of Internet of Things (IoT) deployments has created new opportunities to provide diverse data to applications. By leveraging this IoT data, applications can achieve real-time situation awareness in domains such as healthcare, robotics, and disaster management. While cloud computing supports the acquisition, processing, and dissemination of IoT data, relying solely on cloud-based solutions can result in high latency, and lead to privacy related issues. Introducing an edge computing layer between IoT devices and the cloud helps address these issues by bringing processing and storage closer to the data sources, while offloading computationally intensive or long-term storage tasks to the cloud. This approach to designing storage and processing components is known as the computing continuum. Existing computing continuum frameworks are often domain-specific and difficult to adapt for seamless data acquisition from heterogeneous sources to serve multiple applications. To address this limitation, this paper proposes Q-PRIME (Quality- and Privacy-aware Management for Edge–Cloud IoT), a platform that dynamically acquires and delivers IoT data with awareness of both quality and privacy requirements by intelligently distributing processing across edge and cloud layers, while also providing storage and visualisation services. The effectiveness of Q-PRIME in enabling quality- and privacy-aware situational awareness is evaluated based on Quality of Service (QoS) and privacy preservation metrics. Q-PRIME demonstrates up to 2.9× lower latency at the edge, consistent system performance, accurate quality-aware storage decisions, and negligible exposure of sensitive data.

Edge computing reduces latency by processing data nearer to its origin, improving real-time decision-making under applications like smart cities and autonomous systems that demand quick reactions. However, conventional edge architectures suffer from significant issues such as power consumption, computation, and security. In order to overcome these limitations, this study combines the quantum-based reversible logic in edge computing by designing a reversible arithmetic and logic unit (ALU) implemented with IBM Quantum Laboratory. The proposed reversible ALU uses a Feynman gate (FG), a Reversible Multiplexer (RMUX1) gate, and an NMG6 gate to guarantee the logical reversibility and the preservation of information with the least dissipation of energy. The circuit is simulated and verified using the IBM Quantum laboratory. Its performance is evaluated using quantum cost (QC), garbage output (GO), constant input (CI) and number of gates (GC) metrics. In addition, simulation results together with a literature-based comparison indicate that the proposed reversible ALU achieves favorable QC, GO, CI, and GC metrics relative to selected earlier reversible ALU designs. These circuit-level results indicate the potential of reversible quantum design principles for low-power and efficient computation in future edge-oriented architectures. The presented circuit-level results indicate the potential of reversible quantum design principles for low-power computation in future edge-oriented architectures.

As Edge Intelligence (EI) becomes increasingly prevalent in domains such as smart healthcare, manufacturing, and critical infrastructure, ensuring data privacy while maintaining system efficiency is a growing challenge. This paper presents a new privacy-preserving machine learning (PPML) framework tailored for EI applications, including a four-layer system architecture and training and inference algorithms. We focus on three leading approaches: Differential Privacy (DP), Secure Multi-party Computation (SMC), and Fully Homomorphic Encryption (FHE), and assess their impact on key performance metrics, including model accuracy, response time, and energy consumption. Results from real implementation and extensive trace-based simulations of inference tasks show that DP generally preserves throughput and latency close to plaintext baselines, while accuracy drops with model complexity (up to 35 percent on AlexNet and under 18 percent on LeNet for FordA). SMC performance is driven by communication; network bandwidth and round complexity determine end-to-end latency. For AlexNet, increasing link capacity from 250 Mbps to 500 Mbps reduces latency by about 30 percent. FHE is highly sensitive to model structure and numerical precision bit width, with tighter parameters imposing substantial compute overhead; we observe roughly a 1000 times increase in response time compared to DP. Beyond efficiency, DP shifts the privacy-utility-extractability frontier by reducing the attacker's data efficiency in black-box model stealing, whereas SMC and FHE, while protecting inputs and parameters during inference, require complementary output controls to achieve similar resistance to extraction. These findings provide critical insights into the trade-offs between privacy, performance, and resource efficiency in edge computing scenarios.

Resource management and task distribution in real-time have become increasingly challenging due to the growing use of latency-critical applications across dispersed edge-cloud infrastructures. Intelligent adaptable mechanisms capable of functioning effectively on resource-constrained edge devices and responding quickly to dynamic workload changes are required in these situations. In this work, we offer a learning-based system for autonomous resource allocation across the edge–cloud continuum that is both lightweight and scalable. Two models are presented: TinyDT, a small offline decision tree trained on state-action information retrieved from an adaptive baseline, and TinyXCS, an online rule-based classifier system that can adjust to runtime conditions. Both models are designed to operate on resource-constrained edge devices while minimizing memory overhead and inference latency. Our analysis demonstrates that TinyXCS and TinyDT outperform existing online and offline baselines in terms of throughput and latency, providing a reliable, power-efficient solution for next-generation edge intelligence.

Remote Patient Monitoring Systems (RPMS) require efficient resource management to prioritize life-critical data in latency-sensitive healthcare environments. This research introduces an Integrated Queuing and Certainty Factor Theory (IQCT) model aimed at optimizing bandwidth allocation and task scheduling within fog-edge-cloud architectures. IQCT prioritizes patient requests in real time by classifying them into emergency, warning, and normal categories using certainty factor(CF) -based urgency assessment. Simulated on Raspberry Pi fog nodes with the UCI Heart Disease dataset, its performance was benchmarked against FCFS, PQ, and WFQ using metrics such as latency, energy consumption, and response time under varying workloads. IQCT reduced latency for emergency requests by 54.5% and improved network efficiency by 30.08% compared to FCFS. It also lowered response and execution times by 49.5% and 36%, and decreased fog-layer energy consumption by 30.8%. Scalability tests confirmed stable quality of service (QoS) under peak loads, demonstrating adaptability to dynamic demand. The adaptation of PQ and CF theory can lead to more efficient and optimized performance in RPMS. The IQCT model has significantly reduced the latency by 54.5% in emergency situations, in comparison with the existing models.

In the era of diverse hardware platforms, a critical element of neural network architecture optimization is ensuring that the architecture is optimized for a specific deployment environment. This paper aims to present an effective distributed method for optimizing neural architectures to address the challenge of achieving multiple objectives simultaneously. Using the power of distributed computing, our method employs an Actor-Critic framework to explore the architecture search space while considering multiple resource-related objectives such as memory footprint for deployments in memory-constrained systems, and the amount of computation required given a number of Floating-point Operations Per Second (FLOPs) for applications deployed on energyconstrained devices. An important innovation that is proposed in this research is the integration of hardware awareness, multi-objective optimization, and multi-modality in data. It enables neural architectures to meet strict hardware constraints, including memory efficiency and inference speed, while handling a wide range of data modalities. The proposed method was evaluated in four scenarios, incorporating 3 image, 4 text, 8 graph, and 8 time series benchmarks. Despite the diversity of data modalities, the method demonstrates consistent performance and adaptability.

In recent years, edge computing has emerged as a promising solution for deploying IoT applications that demand minimal latency. By leveraging Function as a Service (FaaS) at the edge, it is possible to achieve efficient and scalable computing capabilities. However, implementing serverless deployment at the edge presents challenges such as auto-scaling, resource management, and mitigating cold-start delays, particularly due to the limited resources available. These challenges are even more significant in workflow-based applications, where tasks are interdependent. This article introduces a dynamic approach for executing serverless workflows at the edge, consisting of three key components: initial function placement, request scheduling, and dynamic adjustment. The initial placement leverages the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to deploy function instances across edge nodes. Request scheduling, on the other hand, distributes requests among these instances using a pattern graph matching algorithm. Finally, the dynamic adjustment component periodically refines placement and scheduling strategies to adapt to changing demands, utilizing a local search technique known as simulated annealing. Evaluation results indicate that the proposed solution reduces the average makespan of workflows by up to 86% compared to state-of-the-art methods.

Federated Learning (FL) enables collaborative model training across various distributed devices without sharing raw data. Client failures, variable energy availability, and outages of edge servers contribute to unreliable training participation, incomplete model updates, and failures at the system level during the aggregation process. In this study, we introduce a reliability-aware workload allocation FL framework (FedRAW) aimed at improving system reliability in failure-prone edge computing systems. Our approach dynamically modifies client workloads based on their failure history and integrates a lightweight backup mechanism to maintain aggregation continuity during edge server failures by backup servers handling. Additionally, we employ Bayesian optimization to fine-tune workload parameters, achieving improved energy efficiency. Experimental results reveal that our proposed method improves model accuracy while reducing energy consumption compared to recent federated learning algorithms.

Edge Federated learning represents a novel architecture at the intersection of federated learning, edge computing, and cloud computing. Edge Federated learning enables efficient and scalable machine learning model training at the network edge by preserving local training data privacy. The performance of this architecture is measured in terms of model accuracy, training time, and communication cost, which vary with the quality and availability of the underlying hosted network infrastructure and computing resources. This paper addresses the challenges related to network infrastructure and systematically characterizes network dynamics such as latency, packet loss, and bandwidth constraints in edge-federated learning environments and quantifies their impact on model training time and accuracy. Furthermore, this paper establishes linear regression equations to predict training time, using regression analysis on quantified data to inform the model. The results showed that high latency, low bandwidth, and high packet drop do not affect accuracy. Additionally, high latency and packet drop have a significant impact on model training time, whereas low bandwidth has a minimal impact.

The placement of microservices in the compute continuum plays a vital role in delivering services that comply with customers’ needs, such as reduced latency, storage requirements, quality of service and availability. To achieve customers’ needs in the geographically dispersed architecture of the compute continuum, Service Level Objectives (SLOs) have been largely used in decision-making to place microservices. However, because low-level SLOs increase the barrier to entry for continuum users, placement decisions based on high-level business vocabulary are required if the compute continuum is to be adopted at scale. This paper proposes an architecture for microservices placement decisions in the computing continuum utilizing highlevel user intents described in natural language as input. The approach utilizes Generative Artificial Intelligence to translate the intents to low-level SLOs, which are used along with the infrastructure description to decide where different microservices that compose an application must be deployed so that SLOs are met. We implement and evaluate a prototype of the architecture to demonstrate the approach’s feasibility.

Serverless computing is a cloud computing model that abstracts server management, allowing developers to focus solely on writing code without concerns about the underlying infrastructure. This paradigm shift is transforming application development by reducing time to market, lowering costs, and enhancing scalability. In serverless computing, functions are event-driven and automatically scale in response to events such as data changes or user requests. Despite its advantages, serverless computing presents several research challenges, including managing state for ephemeral functions, mitigating cold start delays, optimizing function composition, debugging, efficient auto-scaling, resource management, and ensuring security and compliance. This special issue focused on addressing these challenges by promoting research on innovative solutions and exploring the potential of serverless computing in new application domains.

The Internet of Things (IoT) encompasses diverse characteristics, such as varying load frequencies and performance requirements. Designing static IoT systems to cover such varying loads is challenging. Self-adaptive systems, enhanced by artificial intelligence, can offer better performance by responding dynamically to changing conditions. Empirical research is essential to validate such systems. In this paper, we contribute by designing experiments to assess the performance of multiple IoT architectures under various load frequencies. Using the empirical data collected, we train an artificial neural network to predict response times for untested frequencies, identifying optimal scenarios for self-adaptive architecture transitions. We present a dataset of 2,641,008 points regarding the response times of requests for several industrial IoT deployment architectures. We perform an extensive systematic evaluation of 4,374 cases indicating 29.6% improvements in terms of reducing mean response times. Additionally, we provide prototypical tool support for practical implementations and to make our approach easy-to-use.