Unmanned Aerial Vehicles (UAVs) have recently emerged as a promising platform for acquiring high-resolution imagery in urban environments. Efficiently detecting cars from these images is vital for various applications, including traffic management, urban planning, and security. However, the abundance of features and the large variability in the appearance of cars within high-resolution UAV images pose significant challenges. This paper introduces a novel multi-task framework designed to enhance car detection by focusing specifically on road regions. The model utilizes a shared encoder and a fully convolutional network decoder, augmented by an attentive binary fusion for road and car segmentation. For car detection, we combine a deep layer aggregation with a CenterNet detection head. During training, UAV images are downsampled, passed through the encoder and both decoders, generating road/car confidence maps and car detection results. In the inference phase, a region extraction module is designed to extract high-resolution road segments according to the road segmentation mask. To enhance detection accuracy, the region extraction module concatenates the input image with the car confidence map. We also introduce a scale-weighted focal loss in response to challenges associated with detecting smaller cars in high-resolution UAV images. Experimental results on the UAVid2020 and VisDrone2020 datasets demonstrate the superiority of our model in both inference time and accuracy, meeting the real-time requirements of intelligent traffic systems.

Most conventional Federated Learning (FL) models are using a star network topology where all users aggregate their local models at a single server (e.g., a cloud server). That causes significant overhead in terms of both communications and computing at the server, delaying the training process, especially for large scale FL systems with straggling nodes. This article proposes a novel edge network architecture that enables decentralizing the model aggregation process at the server, thereby significantly reducing the training delay for the whole FL network. Specifically, we design a highly-effective in-network computation framework (INC) consisting of a user scheduling mechanism, an in-network aggregation process (INA) which is designed for both primal- and primal-dual methods in distributed machine learning problems, and a network routing algorithm with theoretical performance bounds. The in-network aggregation process, which is implemented at edge nodes and cloud node, can adapt two typical methods to allow edge networks to effectively solve the distributed machine learning problems. Under the proposed INA, we then formulate a joint routing and resource optimization problem, aiming to minimize the aggregation latency. The problem turns out to be NP-hard, and thus we propose a polynomial time routing algorithm which can achieve near optimal performance with a theoretical bound. Simulation results showed that the proposed algorithm can achieve more than 99$\%$of the optimal solution and reduce the FL training latency, up to 5.6 times w.r.t other baselines. The proposed INC framework can not only help reduce the FL training latency but also significantly decrease cloud's traffic and computing overhead. By embedding the computing/aggregation tasks at the edge nodes and leveraging the multi-layer edge-network architecture, the INC framework can liberate FL from the star topology to enable large-scale FL.

Fog computing is one of the promising technologies for realizing global-scale Internet of Things (IoT) applications as it allows moving compute and storage resources closer to IoT devices, where data is generated, in order to solve the limitations in cloud-based technologies such as communication delay, network load, energy consumption, and operational cost. However, this technology is still in its infancy stage containing essential research challenges. For instance, what is a suitable fog computing scheme where effective service provision models can be deployed is still an open question. This paper proposes a novel multitier fog computing architecture that supports IoT service provisioning. Concretely, a solid service placement mechanism that optimizes service decentralization on fog landscape leveraging context-aware information such as location, response time, and resource consumption of services has been devised. The proposed approach optimally utilizes virtual resources available on the network edges to improve the performance of IoT services in terms of response time, energy, and cost reduction. The experimental results from both simulated data and use cases from service deployments in real-world applications, namely, the intelligent transportation system (ITS) in Ho Chi Minh City, show the effectiveness of the proposed solution in terms of maximizing fog device utilization while reducing latency, energy consumption, network load, and operational cost. The results confirm the robustness of the proposed scheme revealing its capability to maximize the IoT potential.

This article introduces a dynamic cloud-based marketplace of near-realtime human sensing data (MARSA) for different stakeholders to sell and buy near-realtime data. MARSA is designed for environments where information technology (IT) infrastructures are not well developed but the need to gather and sell near-realtime data is great. To this end, we present techniques for selecting data types and managing data contracts based on different cost models, quality of data, and data rights. We design our MARSA platform by leveraging different data transferring solutions to enable an open and scalable communication mechanism between sellers (data providers) and buyers (data consumers). To evaluate MARSA, we carry out several experiments with the near-realtime transportation data provided by people in Ho Chi Minh City, Vietnam, and simulated scenarios in multicloud environments.

Cloud computing based Data-as-a-Service (DaaS) has become popular. Several data assets have been released in DaaSes across different cloud platforms. Nevertheless, there are no well-defined ways to describe DaaSes and their associated data assets. On the one hand, existing DaaS providers simply use HTML documents to describe their service. This simple way of service description requires user to manually perform service lookup by reading the HTML documents to understand DaaSes as well as their provided data assets. On the other hand, existing service description techniques are not suitable for describing DaaSes because they consider only service information. The lack of well-structured/linked model to describe DaaSes hinders the automatic service lookup for DaaSes and the integration of DaaSes into data composition and analytic tools. In this paper, we propose DEMODS, a Description Model for DaaS, which introduces a general linked model to cover all basic information of a DaaS. Besides the basic DaaS description model, we also introduce an extended model that integrates existing work in describing quality of data, data and service contract, data dependency, and Quality of Service (QoS). We present a mechanism to incorporate DEMODS into both new and existing DaaSes. Finally, a prototype of DEMODS has been developed to evaluate the effectiveness of the proposed model.

With diverse and rich offerings from cloud computing providers in the open cloud market, scientists have great opportunities to design and conduct complex applications by utilizing and combining computational resources, software components and data sources in elastic manners. While existing techniques focus mainly on resource elasticity in single cloud infrastructure, scientists expect to design their applications being elastic in multiple dimensions to ensure that they applications can operate on multiple clouds with minimum software engineering effort. In this paper we will focus on providing techniques for scientists to compose elastic high performance applications by utilizing traditional software components, user-provided components and cloud services. We characterize elastic compositions via their resource, quality, cost, available time and usage right elasticity, thus enabling scientists to evaluate and decide how to develop, deploy and control the compositions to match their elastic needs. To illustrate our approach, we will present several real-world application compositions for multi-cloud environments.

In recent years, huge amount of digital documents have been created in online social communities. Locating and recommending relevant documents to users in need is not a trivial task. In this paper, we introduce an approach to computing document similarity for building document recommender systems. This approach takes into account document internal content, together with social tags and associated users. These three factors are considered as three dimensions of a document in social space. The similarity computation is performed independently in each dimension, and then aggregated according to user preferences. This allows users to customize their views over the document. Experiments have been conducted with the proposed approach using data crawled from Cite Like. Experimental result has confirmed that all three dimensions have important contributions to document similarity computation.

Despite the success High Performance Computing (HPC) across a number of application domains, the adoption of HPC resources and applications is still limited, primarily due to its high capital cost, system complexity, application availability, and service delivery model. Recently, several research efforts have shown that the emerging Cloud Com-puting service model can improve on-demand access to HPC capacity as utility. This paper introduces a framework for on-demand composing and deploying available HPC applica-tions as services on HPC clouds. The composition is enabled by an ontology that describes dependencies and relationships among HPC software and resources.

This paper proposes an application of the Collaborative e-Science Architecture (CeSA) to enable e-Research in combustion research community. A major problem of the community is that data required for constructing modelling might already exist but scattered and improperly evaluated. That makes the collection of data for constructing models difficult and time-consuming. The decentralised P2P collaborative environment of the CeSA is well suited to solve this distributed problem. It opens up access to scattered data and turns them to valuable resources. Other issues of the community addressed here are the needs for computational resources, storages and interoperability amongst different data formats can also be addressed by the use of Grid environment in the CeSA.