Autonomous vehicles face perception challenges due to occlusions, limited sensor ranges, and adverse weather. Vehicle-to-Everything (V2X) cooperative perception mitigates these limitations by enabling vehicles to share sensor data. However, existing methods rely on supervised learning, requiring costly manual 3D annotations, exhibiting limited generalization, and employing static fusion strategies that fail under communication disruptions. We propose V2X-JEPA, a self-supervised framework that extends the joint-embedding predictive architecture (JEPA) to V2X cooperative perception for both Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure scenarios. V2X-JEPA learns semantic representations via latent embedding prediction, eliminating the need for manual annotations during pretraining. We introduce three cooperative masking strategies and grid spatial attention fusion, which adapt to communication quality and agent reliability. Extensive evaluation on OPV2V and DAIR-V2X benchmarks shows that V2X-JEPA reduces annotation requirements by 85% while achieving competitive performance. On OPV2V (LiDAR, V2V), V2X-JEPA achieves 94.5% [email protected] and 89.2% [email protected]. On DAIR-V2X (camera, V2I), it achieves 24.8% [email protected] and 11.2% [email protected]. It outperforms V2X-ViT by 4.7%, CoCa3D by 2.2%, and CooPre by 2.8%. V2X-JEPA is efficient, with 98M parameters and 78-ms inference time, and demonstrates high robustness, degrading only 5.1% under 20% packet loss, supporting practical deployment in bandwidth-constrained environments.

Federated Learning (FL) is increasingly used in healthcare to enable collaborative model training across decentralized medical institutions while preserving patient privacy. Despite its promise, FL faces significant challenges in medical image processing, such as class imbalance and client data heterogeneity, which can lead to biased models and reduced accuracy in detecting rare diseases. To address these issues, we propose a novel approach, called Syndicated Federated Learning (SyndFL), comprising a multi-layer client selection algorithm that emphasizes fair client representation based on dataset size, learning performance, label distribution, and domain-specific features. SyndFL not only prioritizes clients from minority data clusters but also includes adaptive weighting to ensure that rare conditions receive adequate representation, reducing the risk of bias in model aggregation. Experiments conducted on healthcare image datasets demonstrate that SyndFL achieves a $\mathbf{1 0}-\mathbf{2 0 \%}$ improvement in detecting rare conditions compared to standard FL methods, significantly enhancing both model robustness and fairness in clinical decision-making and diagnostics.

Federated Learning (FL) and Split Learning (SL) are innovative machine learning approaches that enable decentralised model training while ensuring patient data privacy in healthcare. However, their effectiveness is challenged by frugal labelling and label scarcity, which hinder the development of high-quality models due to the limited availability of labelled data. This paper explores methods to overcome these challenges by leveraging semi-supervised learning, transfer learning, collaborative label sharing, and federated semi-supervised learning. These strategies optimise model training by utilising labelled and unlabelled data, reducing the need for extensive manual annotation while maintaining accuracy and compliance with privacy regulations. Secure collaboration techniques, such as multi-party computation and differential privacy, facilitate ethical data-sharing practices. The paper highlights real-world applications and discusses future research directions to enhance FL and SL for more efficient, scalable, and privacy-preserving healthcare AI solutions.

While Federated Learning (FL) manages well data diversity in distrusted learning, it faces additional complexities in scenarios with "frugal labeling," where the client nodes host a mix of partially or fully unlabelled datasets. This paper introduces FedDAFL, a novel approach tailored for label-scarce federated settings, that enables learners with unlabelled datasets to leverage fully labelled teachers, thus incorporating domain adaptation techniques within a multi-teacher multi-learner framework. Through the integration of self-learning and semi-supervised methods, FedDAFL aims to enhance model accuracy using each client’s unlabelled data. Comparative analysis with established techniques like Federated Adversarial Domain Adaptation (FADA) and Federated Knowledge Alignment (FedKA), demonstrates FedDAFL’s superior accuracy across independent and identically or non-identically distributed (non-IID) datasets. Even if the clients possess single class data, FedDAFL adeptly tackles frugal labeling challenges in federated transfer learning, exhibiting competitive performance against existing methods. Results validated on Pneumonia chest X-ray datasets underscore FedDAFL’s adaptability in handling limited labels and non-IID setups, demonstrating its applications to various settings.

This paper proposes a novel selective federated learning (FL) algorithm, called fittest aggregation and slotted training (FedFaSt). It relies on a “free-for-all” client training process to score clients' efficiency while applying the “natural selection” principle to elect the fittest clients to be used in FL training and aggregation processes. While relying on a combined data quality and training performance metric for scoring clients, FedFaSt implements a slotted training model enabling teams of fittest clients to participate in the training and aggregation processes for a fixed number of successive rounds, called slots. Performance validation using X-ray datasets reveals that FedFaSt outperforms selective federated learning algorithms like FedAVG, FedRand, and FedPow in terms of accuracy, convergence to the global optimum, time complexity, and robustness against attacks.

The Cloud computing system is characterized by large scale servers being utilized by an even larger number of users. It is a system where there is the need to frequently and efficiently schedule and manage different application tasks, with varied service requirements. One of the challenges of Cloud computing is managing the quality of service (QoS) rendered to users, specifically scheduling tasks between users and Cloud resources in a timely manner. Cloud users usually have widely diverse QoS requirements and meeting these simultaneously is also a challenge. In this paper, in order to improve on Cloud resource allocation and specifically to tailor it towards meeting varied QoS requirements of users, we proposed a new algorithm which combines Differential Evolution with the Shapley Value economic mode. This combination allows us measure the contribution of each virtual machine (VM), so as to improve the probability of obtaining a better tasks-to-resource allocation thereby improving user satisfaction. From results of conducted experiments, when compared with the traditional DE (Differential Evolution) algorithm and the conventional task-VM binding policy in CloudSim, both for allocations where special QoS requirements are required and in instances of multiple QoS requirements; the modified Shapley value based DE algorithm (SVBDA) shows significant improvement.

Over the years, traffic lights have been used to manage traffic at road intersections. Though relatively effective, the vehicular queues that build up at each intersection being managed by a traffic light and the subsequent delay thereof, could have adverse effects on medical emergency vehicles (EVs). Reports have shown that queues at traffic intersection can increase travel times of medical EVs by an average of 20%. This in many instances could mean the difference between life and death. Prioritizing EVs could be a potential solution to this challenge. In this paper, an Internet of Things based priority preemption model for EVs in smart cities is proposed. It leverages on sensors to dynamically track the EV’s location and speed, and adaptively adjusts the timing sequence of all traffic lights on the EV’s path. This ensures that the EV experiences little or no delay to and from its destination. Experimental results show that the proposed model has the potential to reduce travel delays experienced by medical EVs by up to 35%.

Reportedly starting in Wuhan, China in 2019, the Corona virus 2019 (Covid-19) is a pandemic that has hit many countries of the world. Globally the number of infections and death cases have been on the rise and Africa is no exception. As of May 2020, South Africa had been the most affected country in Africa with Cape Town being the pandemic's epicentre. Preempting the pandemic rather than attempting to cure infected patients is very crucial to Africa, considering its poorer healthcare system compared to developed countries, which have struggled to curtain the ravaging pandemic despite the advanced state of their healthcare systems. This paper proposes two models that i) validate the proposed protective measures using epidemic modelling; ii) pre-empt the evolution of the pandemic through data analytics. The first model builds around the classic SIR model to simulate the main protective measures suggested by the World Health Organisation; while the second built on regression models to predict future confirmed cases. Real Covid-19 data of the city of Cape Town were used for the simulations and results reveal the accuracy of the models and the relevance of combining simulation modelling and data analytics as relevant tools in the fight against the pandemic.

Interests in the exploration and use of Unmanned Aerial Vehicles (UAVs) for service provision have risen in recent years. However, the use of UAVs to extend or solely provide Fifth Generation (5G) wireless network coverage in rural and low income areas is one application area that is yet to be extensively researched. To this end, this paper proposes a topological design model for building an airborne network to provide coverage in rural areas. The model uses a combination of Low Altitude Platform (LAP) as a base station and a cluster of UAVs as cellular access points to provide wireless access to ground users. A combination of performance metrics including Signal-to-Noise Ratio, communication range and residual energy of UAVs are used as guide for designing a robust airborne network with multiple sink nodes. Topology relaxation techniques are applied to design these meshed airborne networks respectively called Multi-Sink Airborne Network with Inter-Cluster Communication through the LAP (MSLBACK) and Multi-Sink Airborne Network with an Inter-Cluster Connection through UAV Gateways (MSGBACK). Compared to myopic approaches (that may lead to isolated UAVs), these airborne networks have more economic relevance as they ensure effective utilization of all UAVs in providing 5G connectivity to ground users. Simulation results reveal that MSLBACK and MSGBACK outperform the state-of-the-art algorithms.

Dynamic spectrum markets for future wireless network devices will involve complex spectrum transactions. The spatio-temporal spectrum availability and the existence in the market of many diverse players with different power limit and channel bandwidth requirements, calls for new ways of allocating the spectrum [1]. Due to its perceived fairness and allocation efficiency, an auction strategy has been advocated as the best strategy for dynamic spectrum markets. However, designing an auction mechanism for a dynamic spectrum market is challenging because of the nature of the spectrum involved and the diverse secondary users, i.e., network players, involved, resulting into an auction mechanism with many complementary requirements. Many attempts from researchers result into spectrum auctions that meet only a few of these requirements. In this paper, a three-stage spectrum auction that not only meet most of these requirements but also ensures optimal spectrum allocation, is proposed.

The HL7 standard was initially designed to enable interoperability of healthcare information within large hospitals. HL7's focus has recently been extended to regional healthcare cloud infrastructures but not yet reached the lightweight fog infrastructures emerging from niche healthcare areas such as body area networks (BANs) and the massive flow of personal medical records resulting from off-the-shelf e-health devices and networks. This paper addresses the issue of interoperability between fog and cloud computing platforms by i) proposing a framework for a standardized exchange of information between healthcare entities ii) designing and implementing a software tool to be integrated into medical data dissemination protocols to ensure interoperability and iii) evaluating the impact of the software tool on the transport of data when exchanging healthcare information using “in-band” and “out-band” transport over the IEEE802.15.4/ZigBee and WiFi protocols. Our results reveal that, for the lightweight devices used by fog infrastructures, “out-band” transport over WiFi with edge data translation into HL7 records is a better option than “in-band” transport.

Data muling using UAVs/drones is currently emerging as an alternative to the traditional traffic engineering techniques used in wireless sensor networks, when wireless communication is not an option or the least cost-efficient solution. This paper revisits the issue of traffic engineering in Internet-of-Things (IoT) settings, to assess the relevance of using UAVs for the persistent collection of sensor readings from the sensor nodes located into an environment and their delivery to base stations where further processing is performed. We consider a persistent path planning and UAV allocation model, where a team of UAVs coming from various base stations are used to collect data from ground sensors and deliver the collected information to their closest base stations. This problem is mathematically formalised and proven to be NP-hard. We propose a heuristic solution for the problem and evaluate its relative efficiency through simulation.

The integration of wearable and integrated sensing technologies with cloud computing has the potential to promote the development of large-scale monitoring applications and to provide public safety as a service to end-users. In this paper, we proposed a system that relies on the integration of heterogeneous sensor networks and cloud computing to build distributed public safety databases and to provide cooperative composite services provisioning in infrastructure-less areas. The contributions are three fold. The first contribution consists in designing a Sensor-Clouds based architecture that integrates different WSNs and cloud computing and aims to report on the evolution of a particular set of public safety threats. The second contribution focuses on describing a system model for composite Sensor-Cloud services provisioning that allows to identify and classify the sensor-cloud services that can be made available to end-users regardless of the geographical distribution of sensor networks. The third contribution describes a two-layer composite service delivery in order to efficiently cover several services requested by end-users. While, the first level defines a static composite service delivery the second level defines dynamic composite services that are delivered and versioned based on a set of heuristics to address users' requirements and expectation. The proposed approach is evaluated through a simulation.

The recent advances in sensor/actuator and RFID technologies have spun out a new healthcare model enabling capture and dissemination of patient vital signs over the Internet for ubiquitous monitoring of these patients anytime and from anywhere. This provides new opportunities for enhancing healthcare through participatory consultation, medical diagnosis and many other novel healthcare services. Some of the advantages of this emerging technology referred to in this paper as “Cyber-healthcare” includes low acquisition cost, flexible deployment and improved accuracy resulting from replacing manual operations by fully digitized processes. It is expected that the emerging healthcare technology will change the way healthcare is delivered in both rural and urban settings of the developing world by building upon this technology to leapfrog from poorly prepared to medically equipped environments capable of tackling some of the most challenging medical issues of the developing world such as patients' vital signs capture, patient prioritization and preparedness to virus outbreaks such as Ebola. This paper proposes a Cyber-healthcare system as a first step towards the implementation of least cost digital health systems in the developing countries. We assess the field readiness of the off-the-shelf sensor technology used by the system and evaluate the performance of its underlying patient prioritization module using two machine learning algorithms: 1) multivariate linear regression and 2) support vector machine.

Network management is revisited in the emerging ubiquitous sensor networks (USNs) that form the Internet-of-the-Things (IoT) with the objective of evaluating the impact of traffic engineering on energy efficiency and assessing if routing simplicity translates into scalability. USN management is formulated as a local optimization problem minimizing the number of traffic flows transiting by a node: the nodes traffic flow interference with other nodes. The least interference beaconing algorithm (LIBA) is proposed as an algorithmic solution to the problem, and the least interference beaconing protocol (LIBP) as its protocol implementation. LIBP extends the beaconing process widely used by collection protocols with load balancing to improve the USN energy efficiency. Simulation results reveal the relative efficiency of the resulting traffic engineering scheme compared to state of the art protocols. These results show up to 30% reduction in power consumption compared to TinyOS beaconing (TOB), and up to 40% compared to collection tree protocol (CTP) while sustaining better performance in terms of scalability.

Despite global advancements in technology and inter-trade volumes, Sub-Saharan Africa is the only Region where cases of hunger have increased since 1990. Rampant and frequent droughts are one of the major causes of this. Monumental and mostly donor-funded projects have been mounted to counter this but with little success. One of the latest strategies being experimented is a community-based early warning system that seeks to integrate indigenous knowledge with western climate science. This initiative is informed by the realization that, though crucial, weather forecast information provided by the national meteorological departments has little utilization amongst small-scale farmers. Though having generated promising results, the integration project still faces the challenges of scaling up across communities as well as the lack of micro-level weather data. In this paper, we describe how the adoption of mobile phones and wireless sensor networks technology is being used to address these two challenges. Use of denser wireless sensor networks to collect local weather data and mobile phones to disseminate forecasts brings information closer to the farmers that need it most. To ensure that the non-mystical aspects of indigenous knowledge are portable across communities, language technologies (part of artificial intelligence) are used in the design of our system.

It is widely recognized that white spaces identification is an important milestone for the wide deployment of next generation cognitive wireless networks. However, spectrum holes detection tools used for white spaces discovery are still either in the infancy stage or too expensive to enable massive white spaces exploitation. Building upon cheap hardware equipment, this paper presents experiments conducted in the town of Trieste in Italy to sense the environment and find out which frequencies are not being used in a particular place and time-of-the-day. As a a step towards white spaces exploitation, we believe that our experimental frequency exploration is an important milestone upon which white spaces patterns recognition will be built with the aim of using these patterns in wireless network planning and management.

The popularity and the high processing power of today's smart phones have presented computer scientists with a fertile platform on which to implement grid computing for mobile phones. Such grids will not require much investment since they are designed to make use of `idle' power on already existing phones. This is because most smart phone users only use their phones for a few minutes or a few hours every day and yet, these phones are powered up 24/7. These kinds of grids are most favorable to developing countries where the penetration of mobile phone exceeds other forms of ICTs. Once in place, the grids can then be utilized to run the much-needed applications such e-health, e-education and drought prediction. In this paper, we present MobiGrid, a middleware for mobile phone grid that is part of a larger research project that aims at integrating mobile phones and sensors to come up with a drought predication tool for use in the developing countries. MobiGrid is an API on which distributed applications can be built. Unlike the rest of grid middleware solutions, the uniqueness of our approach lies in the fact that the middleware is for mobile phones environment.

The past three decades have experienced a phenomenal emergence of several wireless networks and technologies. This next generation of wireless networks (4G) will be integrated into one IP-backbone to offer improved services to the user. The features of 4G include: wide coverage, high data rates, seamless roaming and personalisation. This paper presents a personalised handoff decision method to offer personalisation in seamless roaming for the next generation of wireless networks. This is done by assigning profiles to different users with different preferences and using these profiles to offer personalised handoff. The integration of these two important features of 4G networks will provide the end user the ability to choose their own preferred networks while they roam freely between heterogeneous networks.

The development of a wireless sensor network (WSN) gateway is challenging for sites where limited infrastructures lead to frequent power shortages and and network unreliability. In this paper we presents a low-power, low-cost, 802.15.4 and 802.11 compatible solution which uses open source software to meet local conditions. Using the SunSPOT motes on a system which is mostly platform independent, our system is based on the Fox embedded Linux board and equipped with a USB flash drive and a USB WiFi adapter. The system can be solar-powered, and the results of a solar system design are presented. All the hardware components are available off-the-shelf and are easy to assemble. We conclude that our system is preferred for applications in remote areas, where a stable power supply and a reliable network infrastructure are lacking. Furthermore, it can be used to extend the range of wireless sensor networks by layering a network of long range motes above islands of low range motes.

We present a new traffic engineering (TE) model which is based on QoS rerouting and uses hybrid resilience to improve the recovery performance of multi-layer networks where an MPLS network is layered above an MPlambdaS network. We formulate the rerouting of the LSPs/lambdaSPs as a multi-constrained problem and use its polynomial reduction to find a heuristic solution that can be implemented by standardized constraint-based routing algorithms. This heuristic solution uses a cost-based routing optimization to achieve different network configurations which multiplex/separate bandwidth-aware LSPs/lambdaSPs on the network links. We formulate the resilience upon failure as a multi-objective problem consisting of finding a resilience strategy that minimizes recovery operation time and maximizes the LSP/lambdaSP restorability. A solution to this problem is proposed where a hybrid resilience framework is used to achieve restoration in the MPLS layer to complement path switching in the MPlambdaS layer. We evaluate the performance of the TE model when rerouting the tunnels carrying the traffic offered to a 23- and 31-node networks. Simulation reveals that the hybrid resilience model performs better than classical recovery mechanisms. In terms of restorability, quality of rerouting paths and rerouting stability

This paper addresses the problem of traffic engineering (TE) to evaluate the performance of evolutionary algorithms when used as IP routing optimizers and assess the relevance of using "gene expression programming (GEP)" as a new fine-tuning algorithm in destination- and flow-based TE. We consider a TE scheme where link weights are computed using GEP and used as either fine-tuning parameters in open shortest path first (OSPF ) routing or static routing cost in constraint based routing (CBR ). The resulting OSPF and CBR algorithms are referred to as OSPFgepand CBRgep. The GEP algorithm is based on a hybrid optimisation model where local search complements the global search implemented by classical evolutionary algorithms to improve the genetic individuals fitness through hill-climbing. We apply the newly proposed TE scheme to compute the routing paths for the traffic offered to a 23-, 28- and 30-node test networks under different traffic conditions and differentiated services situations. We evaluate the performance achieved by the OSPFgep, CBRgepalgorithms and OSPFma, a destination-based routing algorithm where OSPF path selection is driven by the link weights computed by a memetic algorithm (MA ). We compare the performance achieved by the OSPFgepalgorithm to the performance of the OSPFmaand OSPF algorithms in a simulated routing environment using NS. We also compare the quality of the paths found by the CBRgepalgorithm to the quality of the paths computed by the constraint shortest path first (CSPF ) algorithm when routing bandwidth-guaranteed tunnels using connection-level simulation. Preliminary results reveal the relative efficiency of (1) the OSPFgepalgorithm compared to both the OSPFmaand OSPF algorithms and (2) the CBRgepalgorithm compared to CSPF routing

Flow-based routing algorithms using a priori knowledge of the ingress-egress pairs to reduce LSP rejection in MPLS networks have recently been proposed as improvements to the destination-based routing model currently deployed in the Internet. These traffic-aware algorithms incur additional complexity which does not necessarily translate into equivalent performance gains. This paper presents a new routing scheme referred to as least interference optimization (LIO) where the online routing process uses the current bandwidth availability and the traffic flow distribution to achieve traffic engineering in IP networks. A least interference optimization algorithm (LIOA) is presented which reduces the interference among competing flows by balancing the number and quantity of flows carried by a link to achieve efficient routing of MPLS bandwidth-guaranteed LSPs. Initial simulation results show that LIOA performs better than several well known routing algorithms such as the minimum hop algorithm (MHA), open shortest path first (OSPF), constraint shortest path first (CSPF) and minimum interference routing algorithm (MIRA) in terms of several performance parameters including the LSP rejection upon congestion, the successful re-routing of LSPs upon single link failure and the ease of implementation.