The Internet of Things (IoT) is set to become one of the key technological developments of our times provided we are able to realize its full potential. The number of objects connected to IoT is expected to reach 50 billion by 2020 due to the massive influx of diverse objects emerging progressively. IoT, hence, is expected to be a major producer of big data. Sharing and collaboration of data and other resources would be the key for enabling sustainable ubiquitous environments, such as smart cities and societies. A timely fusion and analysis of big data, acquired from IoT and other sources, to enable highly efficient, reliable, and accurate decision making and management of ubiquitous environments would be a grand future challenge. Computational intelligence would play a key role in this challenge. A number of surveys exist on data fusion. However, these are mainly focused on specific application areas or classifications. The aim of this paper is to review literature on data fusion for IoT with a particular focus on mathematical methods (including probabilistic methods, artificial intelligence, and theory of belief) and specific IoT environments (distributed, heterogeneous, nonlinear, and object tracking environments). The opportunities and challenges for each of the mathematical methods and environments are given. Future developments, including emerging areas that would intrinsically benefit from data fusion and IoT, autonomous vehicles, deep learning for data fusion, and smart cities, are discussed.

Smart city advancements are driving massive transformations of healthcare, the largest global industry. The drivers include increasing demands for ubiquitous, preventive, and personalized healthcare, to be provided to the public at reduced risks and costs. Mobile cloud computing could potentially meet the future healthcare demands by enabling anytime, anywhere capture and analyses of patients' data. However, network latency, bandwidth, and reliability are among the many challenges hindering the realization of next-generation healthcare. This paper proposes a ubiquitous healthcare framework, UbeHealth, that leverages edge computing, deep learning, big data, high-performance computing (HPC), and the Internet of Things (IoT) to address the aforementioned challenges. The framework enables an enhanced network quality of service using its three main components and four layers. Deep learning, big data, and HPC are used to predict network traffic, which in turn are used by the Cloudlet and network layers to optimize data rates, data caching, and routing decisions. Application protocols of the traffic flows are classified, enabling the network layer to meet applications' communication requirements better and to detect malicious traffic and anomalous data. Clustering is used to identify the different kinds of data originating from the same application protocols. A proof of concept UbeHealth system has been developed based on the framework. A detailed literature review is used to capture the design requirements for the proposed system. The system is described in detail including the algorithmic implementation of the three components and four layers. Three widely used data sets are used to evaluate the UbeHealth system.

Internet of Things (IoT) is set to revolutionize all aspects of our lives. The number of objects connected to IoT is expected to reach 50 billion by 2020, giving rise to an enormous amounts of valuable data. The data collected from the IoT devices will be used to understand and control complex environments around us, enabling better decision making, greater automation, higher efficiencies, productivity, accuracy, and wealth generation. Data mining and other artificial intelligence methods would play a critical role in creating smarter IoTs, albeit with many challenges. In this paper, we examine the applicability of eight well-known data mining algorithms for IoT data. These include, among others, the deep learning artificial neural networks (DLANNs), which build a feed forward multi-layer artificial neural network (ANN) for modelling high-level data abstractions. Our preliminary results on three real IoT datasets show that C4.5 and C5.0 have better accuracy, are memory efficient and have relatively higher processing speeds. ANNs and DLANNs can provide highly accurate results but are computationally expensive

The education industry around the globe is undergoing major transformations. Organizations, such as Coursera are advancing new business models for education. A number of major industries have dropped degrees from the job requirements. While the economics of higher education institutions are under threat in a continuing gloomy global economy, digital and lifelong learners are increasingly demanding new teaching and learning paradigms from educational institutions. There is an urgent need to transform teaching and learning landscape in order to drive global economic growth. The use of distance eTeaching and eLearning (DTL) is on the rise among digital natives alongside our evolution toward smart societies. However, the DTL systems today lack the necessary sophistication due to several challenges including data analysis and management, learner-system interactivity, system cognition, resource planning, agility, and scalability. This paper proposes a personalised Ubiquitous eTeaching & eLearning (UTiLearn) framework that leverages Internet of Things, big data, supercomputing, and deep learning to provide enhanced development, management, and delivery of teaching and learning in smart society settings. A proof of concept UTiLearn system has been developed based on the framework. A detailed design, implementation, and evaluation of the UTiLearn system, including its five components, are provided using 11 widely used datasets.

Road transportation is the backbone of modern economies, albeit it annually costs $1.25$ million deaths and trillions of dollars to the global economy, and damages public health and the environment. Deep learning is among the leading-edge methods used for transportation-related predictions, however, the existing works are in their infancy, and fall short in multiple respects, including the use of datasets with limited sizes and scopes, and insufficient depth of the deep learning studies. This paper provides a novel and comprehensive approach toward large-scale, faster, and real-time traffic prediction by bringing four complementary cutting-edge technologies together: big data, deep learning, in-memory computing, and Graphics Processing Units (GPUs). We trained deep networks using over 11 years of data provided by the California Department of Transportation (Caltrans), the largest dataset that has been used in deep learning studies. Several combinations of the input attributes of the data along with various network configurations of the deep learning models were investigated for training and prediction purposes. The use of the pre-trained model for real-time prediction was explored. The paper contributes novel deep learning models, algorithms, implementation, analytics methodology, and software tool for smart cities, big data, high performance computing, and their convergence.

Domain adaptation (DA) aims to accomplish tasks on unlabeled target data by learning and transferring knowledge from related source domains. In order to learn a discriminative and domain-invariant model, a critical step is to align source and target data well and thus reduce their distribution divergence. But existing DA methods mainly align the global feature distributions in distorted original space, which neglects their fine-grained local information and intrinsic geometrical structures. Moreover, some methods rely heavily on pseudo-labels to align features, which may undermine adaptation performance and lead to negative transfer. We propose an efficient discriminative manifold distribution alignment (DMDA) approach, which improves feature transferability by aligning both global and local distributions and refines a discriminative model by learning geometrical structures in manifold space. In addition, when learning geometrical structures, DMDA is exempt from the uncertainty and error brought by pseudo-labels of a target domain. It is very concise and efficient to be implemented by integrating learning steps and obtaining solutions directly. Extensive experiments on 68 DA tasks from seven benchmarks and subsequent analyses show that DMDA outperforms the compared methods in both classification accuracy and time efficiency, thus representing a significant advance in the DA field.