Praveen Kumar Donta

Background: With the rapid proliferation of 5G and the Internet of Things, ensuring low latency in edge computing has become crucial for real‐time processing applications. However, existing task scheduling approaches often struggle to balance multiple optimization objectives effectively due to manual parameter tuning and slow convergence to optimal performance. Objective: To address these challenges, we propose a novel Large Language Model (LLM)‐driven edge task scheduling approach, LLM4ETS, which leverages evolutionary computation and auto‐heuristic methods to dynamically evolve towards optimal scheduling strategies. Method: The proposed approach features three core components: (i) a micro‐parameter heuristic approach that minimizes the complexity of parameter settings, (ii) an automated heuristic design process utilizing LLMs for generating and refining scheduling strategies, and (iii) a task sequence modeling mechanism that ensures efficient and adaptive task allocation. These three components work synergistically to optimize resource utilization and reduce latency in edge computing environments. Results: We extensively evaluate the presented approach across multiple real‐world datasets, including the Google Cluster Trace, Alibaba Cluster Trace, and EUA datasets. The experimental results demonstrate that the proposed LLM4ETS approach significantly outperforms state‐of‐the‐art methods in terms of resource utilization and task execution time, confirming its effectiveness and efficiency in edge computing environments.

This special issue explores emerging trends and technologies in ubiquitous network intelligence for next‐generation mobile edge network management and aims to provide valuable insights into the future of intelligent network management in an increasingly interconnected world. We received 48 original research contributions from different geographies, focusing on key challenges such as efficient data gathering from IoT and sensor nodes, distributed learning algorithms for network management, advanced Kubernetes orchestrators for inference, and security in 5G/6G networks. After a rigorous review process, involving three to seven experts per paper and oversight from guest editors, we selected 13 high‐quality research papers, representing a 27% acceptance rate. These papers collectively address the challenges and opportunities in realizing ubiquitous network intelligence from two crucial perspectives: AI for networking and networking for AI. Further, we summarize several open challenges that need to be addressed in the near future related to intelligent ubiquitous mobile edge network management.

The computing continuum (CC) integrates different layers of processing infrastructure, from edge to cloud, to optimize service quality through ubiquitous and reliable computation. Compared to central architectures, however, heterogeneous and dynamic infrastructure increases the complexity for service orchestration. To guide research, this article first summarizes structural problems of the CC, and then, envisions an ideal solution for autonomous service orchestration across the CC. As one instantiation, we show how active inference—a concept from neuroscience—can support self-organizing services in continuously interpreting their environment to optimize service quality. Still, we conclude that no existing solution achieves our vision, but that research on service orchestration faces several structural challenges. Most notably: to provide standardized simulation and evaluation environments for comparing the performance of orchestration mechanisms. Together, the challenges outline a research roadmap toward resilient and scalable service orchestration in the CC.

IoT devices deployed across computing continuum infrastructures present significant security challenges due to resource constraints and decentralization. Traditional centralized intrusion detection systems struggle in such environments because of limited connectivity, high latency, and single points of failure. To address these challenges, this article extends a learning-driven Zero Trust framework tailored to resource-constrained edge environments, with an approach for evaluating lightweight intrusion detection models in Zero Trust edge environments. Our extended approach enables systematic evaluation of lightweight machine learning models for localized intrusion detection, comprising three layers: (i) compilation, (ii) execution, and (iii) measurement. The proposed approach is implemented using Rust and WebAssembly to ensure portable, efficient, and isolated execution across heterogeneous devices. Using this framework, seven representative intrusion detection models (i.e., Decision Tree (DT), Random Forest (RF), k-Nearest Neighbor (KNN), Logistic Regression (LR), Artificial Neural Network (ANN), and Convolutional Neural Network (CNN) variants) were implemented and evaluated on the UNSW-NB15 dataset. Results show that RF achieved the best trade-off between detection accuracy and efficiency, while simpler models (DT and LR) offered near-instant inference with minimal resource usage, making them ideal for highly constrained devices. In contrast, more complex models such as deep neural networks and KNN introduced significant overhead for only modest accuracy gains. These findings underscore the need to balance accuracy and resource efficiency for effective Zero Trust edge security.

Internet of Things (IoT) networks often face conflicting routing goals such as maximizing packet delivery, minimizing delay, and conserving limited battery energy. These priorities can also change dynamically: for example, an emergency alert requires high reliability, while routine monitoring prioritizes energy efficiency to prolong network lifetime. Existing works, including many deep reinforcement learning (RL) approaches, are typically centralized and assume static objectives, making them slow to adapt when preferences shift. We propose a dynamic and fully distributed multiobjective Q-learning routing algorithm that learns multiple per-preference Q-tables in parallel and introduces a novel greedy interpolation policy (GIP) to act near-optimally for unseen preferences. The algorithm learns to optimize for energy efficiency, packet delivery ratio (PDR), and the composite reward, adapting to changing tradeoffs between these metrics without retraining or centralized control. A theoretical analysis further shows that the optimal value function is Lipschitz continuous in the preference parameter, ensuring that the proposed GIP yields provably near-optimal behavior. Simulation results show that our approach adapts in real time to shifting priorities and achieves up to 80%–90% lower energy consumption and up to $5\times $ higher cumulative rewards and packet delivery compared to six baseline protocols, under dynamic and distributed settings. Sensitivity analysis across varying preference window lengths confirms that the proposed DPQ framework consistently achieves higher composite reward than all baseline methods, demonstrating robustness to changes in operating conditions.