Traditional terrestrial computing facilities, i.e., Edge Computing (EC) and Cloud Computing (CC), have limited applicability for remote Internet of Things (IoT) applications due to connectivity challenges. These challenges can be addressed through a Space-Edge Cloud continuum that leverages computing facilities enabled by Low Earth Orbit (LEO) satellites and terrestrial cloud facilities. We aim to minimize the joint latency and energy cost for processing remote IoT data in this continuum by formulating a complex optimization problem. Specifically, our goal is to optimize the complete offloading process by jointly selecting optimal routes, computation nodes, and buffering policies. We develop a Multi-Agent Hierarchical Reinforcement Learning (MA-HRL) solution to solve this problem. The proposed solution is developed in a Python environment and compared with several benchmark methods, demonstrating significant performance gains in terms of reduced latency and energy demands.

Federated Split Learning (FSL) has emerged as a transformative paradigm that synergizes the parallel processing and scalability of Federated Learning (FL) with the computational efficiency and enhanced privacy of Split Learning (SL). This comprehensive survey provides a systematic exploration of FSL, beginning with a detailed taxonomy of Distributed Machine Learning (DML) paradigms, tracing the progression from foundational concepts to advanced frameworks such as Federated Split Transfer Learning (FSTL) and Generalized Federated Split Transfer Learning (GFSTL). It then delves into the core challenges inherent to FSL, including privacy and security risks, system and data heterogeneity, computational and system constraints, communication overhead, and model optimization complexities. The heart of the survey presents a detailed categorization and analysis of FSL's diverse applications across key domains, including the Internet of Things (IoT) and Edge Computing (EC), wireless networks, healthcare, vehicular networks, Large Language Models (LLMs), and Earth Observation (EO). To ground this research, the survey further discusses standardized evaluation methodologies and implementation frameworks, followed by a quantitative visualization of survey data and research trends. The work concludes by synthesizing critical research gaps and outlining promising future directions. By synthesizing insights from over 100 recent research articles and providing a critical analysis of evaluation methodologies, this survey offers an essential roadmap for researchers and practitioners developing scalable, efficient, and privacy-aware distributed intelligence for the 6G and AI era.

The Internet of Things (IoT) is one of the most important applications in a distributed edge-cloud continuum. In this study, we aim to solve the data offloading and buffering policy selection problem for the remote IoT devices to process the data over the space edge cloud continuum formed by the non-terrestrial Edge Computing (EC) facilities, deployed over Low Earth Orbit (LEO) satellites, and the terrestrial cloud infrastructure. In particular, we have developed a constrained optimization problem to minimize the joint latency and energy costs associated with the data processing operation over LEO satellite nodes and cloud facilities. The problem is then decomposed into a hierarchy of decision-making subproblems and addressed using a Hierarchical Reinforcement Learning (HRL) approach. The proposed HRL solution is then implemented within a Python-based simulator, and its performance is compared with several other benchmark solutions. Performance analysis indicates the gain in terms of latency, energy, and resource utilization with adaptive utilization of space computing resources based on the devices’ demands.

Distributed Machine Learning (DML) methods are expected to play a crucial role in the forthcoming 6G era, with the goal of enabling ubiquitous connected intelligence. Distributed intelligence enabled through distributed computing environments, 6G technology, and big data can be extremely supportive of achieving the goals of emerging intelligent IoT applications in the proximity of end users. With this in mind, we propose an advanced Federated Learning (FL) approach for efficiently enabling intelligent applications over latency-critical networks in Non-Terrestrial environments. In the proposed solution, the client and server nodes reduce idle time using a parallel processing approach with the help of a replica of the training model. Next, the proposed FL framework is tested in a Python environment to show its effectiveness with respect to the traditional FL approach.

With recent advancements in Internet of Things (IoT), Machine Learning (ML), and wireless communication networks, e.g., 6G, there has been an increasing interest in Intelligent Vehicular Networks (IVNs), where efficient traffic flow management is one of the most important requirements. In recent times, several new distributed ML methods have been introduced aiming at solving complex networking problems due to their added advantages in terms of learning efficiency and privacy over distributed wireless scenarios. With a focus on the upcoming 6G enabled Intelligent Transportation Systems, real-time traffic flow management is essential for providing the adequate Quality of Service to the end users. Distributed Learning methods can be crucial for solving the traffic management problem in highly dynamic vehicular systems. With this motivation, in this work, we have considered a distributed learning method, belonging to the class of the collaborative Federated Learning (FL) approaches, named Networked FL (NFL) for estimating the dynamic traffic flow over time and space. We have exploited the added advantage of NFL in terms of multi-task FL capabilities for predicting traffic patterns over different times and locations in a service area. The simulation results compared with the traditional centralized FL and independent area-based learning show improvements in terms of learning efficiency, accuracy, and the cost required.

Internet of Vehicles (IoV) is a recently introduced paradigm aiming at extending the Internet of Things (IoT) toward the vehicular scenario in order to cope with its specific requirements. Nowadays, there are several types of vehicles, with different characteristics, requested services, and delivered data types. In order to efficiently manage such heterogeneity, Edge Computing facilities are often deployed in the urban environment, usually co-located with the roadside units (RSUs), for creating what is referenced as vehicular edge computing (VEC). In this article, we consider a joint network selection and computation offloading optimization problem in multiservice VEC environments, aiming at minimizing the overall latency and the consumed energy in an IoV scenario. Two novel collaborative$Q$-learning-based approaches are proposed, where vehicle-to-infrastructure (V2I) and vehicle-to-vehicle (V2V) communication paradigms are exploited, respectively. In the first approach, we define a collaborative$Q$-learning method in which, through V2I communications, several vehicles participate in the training process of a centralized$Q$-agent. In the second approach, by exploiting the V2V communications, each vehicle is made aware of the surrounding environment and the potential offloading neighbors, leading to better decisions in terms of network selection and offloading. In addition to the tabular method, an advanced deep learning-based approach is also used for the action value estimation, allowing to handle more complex vehicular scenarios. Simulation results show that the proposed approaches improve the network performance in terms of latency and consumed energy with respect to some benchmark solutions.

Supported by some of the major revolutionary technologies, such as Internet of Vehicles (IoVs), Edge Computing, and Machine Learning (ML), the traditional Vehicular Networks (VNs) are changing drastically and converging rapidly into one of the most complex, highly intelligent, and advanced networking systems, mostly known as Intelligent Transportation System (ITS). Recently, distributed ML techniques, such as Federated Learning (FL) have gained huge popularity mainly for their advantages in terms of intelligence sharing and privacy concerns. VNs are a natural contender for exploiting FL for solving challenging problems; however, their limited resources, dynamic nature, high speed, and reduced latency requirements often become the bottleneck. V2X communication technologies allow vehicular terminals (VTs) to share their valuable local environment parameters and become aware of their surroundings. Such information can be utilized to build a more sustainable and affordable FL platform for serving VTs. Gaining from recently introduced 3D architectures, integrating terrestrial and aerial edge computing layers, we present here a distributed FL platform able to distribute the FL process on a 3D fashion while reducing the overall communication cost for providing vehicular services. The framework is defined as a constrained optimization problem for reducing the overall FL process cost through a proper network selection between various nodes. We have modeled the FL network selection problem as a sequential decision-making process through a Markov Decision Process (MDP) with time-dependent state transition probabilities. A computation-efficient value iteration algorithm is adapted for solving the MDP. Comparison with various benchmark methods shows the overall improvement in terms of latency, energy, and FL performance.