Computer Networks and Communications

Intelligent Orchestration of Distributed Large Foundation Model Inference at the Edge

Fernando Koch, Aladin Djuhera, Alecio Binotto — Fri, 05 Sep 2025 00:00:00 +0800

Large Foundation Models (LFMs), including multi-modal and generative models, promise to unlock new capabilities for next-generation Edge AI applications. However, performing inference with LFMs in resource-constrained and heterogeneous edge environments, such as Multi-access Edge Computing (MEC), presents significant challenges for workload orchestration due to time-varying network, compute, and storage conditions. In particular, current split inference strategies, which partition LFM layers across nodes, are not designed to adapt to fluctuating workloads, dynamic bandwidth conditions, or evolving privacy constraints in high-utilization MEC environments. In this work, we propose a novel adaptive split inference orchestration framework that elevates both the placement and partitioning of LFM layers to runtime-tunable variables. Specifically, our framework enables real-time, quality-of-service (QoS)-aware management of inference workloads by extending conventional orchestrators with three key services: (1) Capacity-aware workload distribution, which continuously profiles node resources and selects an optimal subset of MEC nodes; (2) Dynamic partition migration, which transparently relocates pre-cut LFM segments in response to changes in utilization or network conditions;(3) Real-time reconfiguration, which dynamically re-splits LFM layers to balance latency, throughput, and privacy. We formalize the joint placement-partitioning problem, outline a reference architecture and algorithmic workflow, and discuss applicability in representative smart city, V2X, and industrial edge scenarios.

Quality of Experience Management in Mobile Networks: Techniques, Constraints, and Emerging Trends for Value-Added Services

Kinge Mbeke Theophane Osee, Valery Nkemeni, Michael Ekonde Sone — Tue, 08 Jul 2025 00:00:00 +0800

The integration of Quality of Experience Management (QoEM) into mobile networks has significantly transformed the telecommunications industry by aligning service delivery more closely with user expectations. This paradigm shift is particularly crucial as user perception of service quality (i.e., perceived QoE) has become a key differentiator in competitive telecom markets, directly impacting overall user satisfaction and retention. This paper presents a comprehensive review of current QoEM techniques, with a particular emphasis on Machine Learning (ML) approaches for predicting user experience and ensuring high-quality service delivery. Additionally, the integration of Software-Defined Networking (SDN) and Network Functions Virtualization (NFV) is highlighted as a key trend in the development of user centric management systems, enabling dynamic network adjustments based on real-time QoE feedback. However, despite these advancements, the transition from traditional Quality of Service (QoS) metrics to QoE-aware frameworks presents significant challenges. These include the complexity of balancing resource allocation across diverse services to maintain optimal user experiences, as well as technical constraints in real-time QoE monitoring. Furthermore, as demand for high-definition streaming and low-latency applications continues to grow, advanced traffic management solutions are becoming increasingly essential. This review also explores emerging trends in Value-Added Service(s) (VAS), particularly within the context of 5G and 6G networks. We conclude this paper by indicating that the effective advancement of QoEM in mobile networks requires interdisciplinary collaboration between academia and industry. Key areas of focus include network architecture, user behaviour analytics, and content delivery mechanisms. A multidisciplinary approach is essential for addressing the complexities of existing QoEM models and ensuring superior user experiences in next-generation networks.

A Secure Prescription System with Machine Learning for SQL Injection Detection

Savina Mariettou, Constantinos Koutsojannis, Vassilis Triantafyllou — Thu, 10 Jul 2025 00:00:00 +0800

This research introduces a secure, web-based prescription system designed to monitor antibiotic consumption and reduce the misuse of critical antibiotics in clinical environments. The system's user interface supports structured documentation and justification of antibiotic use, serving as a clinical surveillance tool that promotes responsible prescribing and contributes to the prevention of hospital-acquired infections through improved antimicrobial stewardship. To ensure robust data protection, the system was evaluated under simulated cyberattacks, including unauthorized access, Denial-of-Service (DoS), Distributed Denial-of-Service (DDoS), and SQL injection attacks. In addition to standard security mechanisms such as Transport Layer Security (TLS) and Elliptic Curve Cryptography (ECC), the system integrates a machine learning–based module implemented in Python to enhance real-time SQL injection detection. The module leverages supervised learning algorithms to classify database queries as malicious or safe, enabling proactive defense against threats targeting sensitive medical records. By embedding machine learning into a secure clinical workflow, the system supports sustainable antibiotic management in hospitals, laying a foundation for scalable, intelligent, and secure e-health infrastructures.

A Review of Virtual Private Network Mobile Application Security

Sat, 11 Oct 2025 00:00:00 +0800

In an era of heightened online risks, the demand for Virtual Private Networks (VPNs) has surged. The VPN market has grown significantly, ranging from popular services like NordVPN, which holds a quarter of the market share, to applications with a small installation base. Studies show that as of 2024, 46% of Americans use at least one VPN application. Given VPNs' role in protecting sensitive data, questions have arisen regarding the security posture of VPN applications themselves. This study systematically reviewed 27 Android VPN applications selected from mobile app stores, following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines to ensure methodological transparency and reproducibility. The analysis utilized application software review, the Mobile Security Framework (MobSF), and public-facing information to assess each application's vulnerabilities, required permissions, and data collection practices. The findings revealed significant variability across the applications and common issues, such as the use of insecure random number generators, excessive permission requests, exported components lacking proper access controls, and the logging of sensitive user information. Based on these results, the study highlights the need for improved secure coding practices to enhance the security posture of existing mobile VPN applications.

Analyzing 5G Heterogeneous Cellular Networks: An Inclusive Examination of Throughput and Energy Efficiency

Maxwell Afriyie Oppong, Emmanuel Ampoma Affum, Kwadwo Ntiamoah-Sarpong — Fri, 21 Nov 2025 00:00:00 +0800

This paper presents an in-depth analysis of energy efficiency in Fifth Generation (5G) cellular networks, with a focus on Heterogeneous Network (HetNet) architectures. The user needs extensive network access in next-generation wireless networks, in addition to the overwhelming demand for high data rates and network capacity. The need for data services that are available anywhere, at any time, requires network operators to construct an increasing number of base stations, which ultimately results in excessive power usage. The 5G network proposes a heterogeneous wireless access network environment as a potential solution to this problem. The goal of green communication was realized with the rise of heterogeneous networks. A heterogeneous network consists of a combination of low-power nodes superimposed over a Macrocell to reduce traffic within the Macrocell and improve cell edge user quality of service. The goal of a heterogeneous network is to reduce energy consumption in mobile wireless networks while simultaneously improving Long-Term Evolution (LTE)-Advanced performance beyond its current limits. Specifically, Microcells, Picocells and Femtocells deployment under the auspices of Macrocell Base Stations (BS). Improved network coverage, increased network capacity, energy efficiency, increased data rates, and better Quality of Service (QoS) are all the outcomes of this networking approach. Two-tier, three-tier, and four-tier network architectures have been used in this article’s energy efficiency analysis using stochastic geometry method. Our analysis considers a network configuration where microcells constitute 30%, picocells 20%, and femtocells 15% of the total deployed base stations, while macrocells remain dominant at 35%. Furthermore, While densification reduces macrocells interference, cross-tier interference increases by up to 20 dB. Multi-tier networks significantly enhance throughput and energy efficiency, but effective interference management is essential to maximize benefits.

Phishing Attack Simulation, Email Header Analysis, and URL Scrutiny: A Comprehensive Approach to Cyber Threat Mitigation

Wed, 02 Jul 2025 00:00:00 +0800

In an era of increasing cyber threats, phishing attacks remain one of the most prevalent and damaging forms of cybercrime. This research aims to simulate phishing campaigns, analyze email headers, and scrutinize URLs to develop a robust framework for identifying and mitigating phishing threats. By leveraging a combination of automated tools and analytical techniques, this study enhances threat detection and response mechanisms within a cybersecurity framework. This research proposes a practical cybersecurity framework integrating open-source tools for phishing simulation, email header analysis, and URL scrutiny to present a comprehensive approach to cybersecurity. The findings contribute to strengthening email security, enhancing threat intelligence, and implementing proactive defense mechanisms, ultimately providing valuable insights into phishing attack methodologies and equipping organizations with the knowledge and tools necessary to mitigate phishing-related risks effectively. This approach stands out by integrating three key analysis strategies—phishing simulation, email header analysis, and URL scrutiny—offering a comprehensive method for identifying and evaluating phishing threats.

Internet of Medical Things: Architecture, Trends, Challenges, and the Evolution Towards IoMT 5.0

Fri, 07 Nov 2025 00:00:00 +0800

The Internet of Medical Things (IoMT) is rapidly reshaping modern healthcare by seamlessly connecting smart medical devices, patients, and care providers through intelligent networks. This paper explores the layered architecture of IoMT, highlighting the roles of sensing devices, secure data transmission, edge and cloud computing, and AI-driven analytics in delivering proactive and personalized medical care. Emerging trends such as remote patient monitoring, smart hospitals, ingestible sensors, and Augmented Reality (AR)/Virtual Reality (VR) applications in IoMT are discussed alongside critical security, privacy, and regulatory challenges. The study further examines innovative use cases and global initiatives that demonstrate the transformative potential of IoMT across diverse healthcare settings. Looking forward, we introduce the vision of IoMT 5.0 through patient-centric design and current evolving technologies such as digital twins, sustainable technologies, decentralized systems, and collaborative robotics which can drive the future of intelligent, resilient, and ethically responsible healthcare ecosystems.

Ensemble Learning-Based DDOS Attack Recognition in IoT Networks

Muhammad Saibtain Raza, Mohammad Nowsin Amin Sheikh, I-Shyan Hwang — Thu, 03 Jul 2025 00:00:00 +0800

According to a quote by Brendan O' Brien," If you think the Internet has changed your life, think again. The Internet of Things is about to change it all over again!". By improving data analytics, IoT operations, and human–machine interaction, the Internet of Things (IoT) and Artificial Intelligence (AI) are coming together to form AIoT, which is transforming modern production in the era of Industry 4.0. Although AIoT promises more sustainability, efficiency, and safety, the increasing use of IoT devices also increases their susceptibility to cyberthreats, among which Distributed Denial-of-Service (DDoS) attacks are among the most common. Unlocking the full potential of AIoT in linked and industrial environments requires addressing these security issues. In this paper, we leverage the publicly available CIC IoT 2023 dataset to conduct a comprehensive analysis of IoT-based cyber threats, focusing on the detection of seven major attack types and their respective subcategories. To guarantee the accuracy and applicability of the input data, a thorough feature extraction procedure was carried out. To evaluate detection performance, we applied a diverse set of six machine learning and deep learning models. Notably, the most successful approach was an ensemble learning strategy, which produced better accuracy and resilience. Thorough validation procedures were used to verify the results' generalizability and dependability, highlighting the promise of advanced learning methods in fortifying AIoT security infrastructures. Our research indicates that ensemble learning and deep learning models are a promising option for implementation in practical AIoT security frameworks as, when appropriately set up, they provide notable benefits for processing and categorizing tabular IoT data.

UAV base-station design method and optimization for urban environment communication with 5G cellular network

Tue, 05 Aug 2025 00:00:00 +0800

During unexpected or temporary events, base transceiver stations (BTSs), also known as base stations (BSs), may be unable to fully meet the flexibility and resilience requirements of upcoming cellular networks. A promising solution to this issue is to serve the cellular networks with low-altitude unmanned aerial vehicle base stations (UAV-BSs), which can assist terrestrial stations in increasing networks’ capacity and coverage. This paper proposes a network planning method for the fast deployment of fifth-generation (5G) cellular networks using a metaheuristic algorithm. This approach aims to determine the minimum number of UAV-BSs that can cover an area the size of a stadium while considering cell capacity, coverage constraints, the system’s spectral efficiency, and the battery life of the UAVs being utilized. We have formulated an optimization problem approach to capture the practical aspects and satisfy the above conditions simultaneously. We have detailed the implementation of a metaheuristic algorithm based on particle swarm optimization (PSO) that finds optimal locations for the UAV-BSs that satisfy all the stadium constraints among various subareas with different user densities. This approach was compared to a genetic algorithm (GA) using the same simulation parameters during performance evaluation. The simulation results indicate that the proposed approach effectively finds the minimum number of UAV-BSs and their 3-D placement so that all users are served based on their traffic requirements. The results also indicate that the quality-of-service (QoS) targets desired for the network are reached in each scenario.