In this paper, the problem of categorizing aircrafts based on their broadcasted ADS-B messages is addressed. Machine learning is proposed for generating models that are able to categorize aircrafts based on their ADS-B messages and flight pattern. The models are trained using real data and tested to evaluate their performance. The results show that high performance is observed especially when balancing the dataset.
In this paper, we present our solution to the Traffic4cast2020 traffic prediction challenge. The information provided includes nine channels where the first eight represent the speed and volume for four different traffic directions, while the last channel indicates the presence of traffic incidents. The expected output should have the first 8 channels of the input at six future timing intervals, while a one-hour duration of past traffic data, in 5mins intervals, are provided as input. We solve the problem using a novel sedenion U-Net neural network. Sedenion networks provide the means for efficient encoding of correlated multimodal datasets. We use 12 of the 15 sedenion imaginary parts for the dynamic inputs and the real part is used for the static input. The sedenion output of the network is used to represent the multimodal traffic predictions. The proposed system achieved a validation MSE of 1.26e-3 and a test MSE of 1.22e-3.
Pedestrian detection is a challenging yet essential task in computer vision with many potential applications in the real world, such as autonomous driving, robotics and surveillance. However, there exit gaps between research and industrial deployment. To bridge these gaps, we perform an extensive evaluation and experiment of the state-of-the-art techniques in a unified framework.
This research project aims to study and analyze the sentiments and emotions of the public in the UAE relevant to online education during and post COVID-19 pandemic. This will provide a faster and more representative insight on the public's mental health during and post the pandemic relevant to online education. This project has the objective of designing and implementing spatiotemporal analytics models that will support the decision makers with detailed insights against a specific aspect during a certain period or at a specific temporal event, and in a specific location. This allows for better understanding of the public's emotional triggers and the impact of the new changes in the educational delivery mode on the mental health of the public. Consequently, informed-intervention plans can be made. we target Twitter as a source of public opinion and emptions due to its popularity in the UAE.
The complex trust domain of the internet entities has become more complex when dealing with Internet of Things (IoT) devices as entities. Trust and reputation systems were designed to establish trust between entities, however, not tailored to the special needs of IoT devices. To provide a general framework for IoT devices in a public network the blockchain technology and reputation systems are used in conjunction. The framework aims to provide a customizable, secure and scalable architecture that enables a service oriented platform for IoT devices and their clients. We provide a design that utilizes subsystems that are motivated with a reward-penalty scheme to aid the end users of the system.
Electrical Impedance Tomography (EIT) is a non-invasive and cheap imaging technology that infers conductivity distributions inside a conductor from surface measurements alone. However, it is infamous for the ill-posed and non-linear nature of its inverse problem which has driven research in machine learning for the implementation of a tactile sensing skin for robotic applications. This paper summarises some of the major contributions in the field, to help track the different approaches to incorporating machine learning for the inverse solution.
Due to the high penetration of renewable energy sources (RESs) into the power grids, future power networks will be more volatile to power oscillations, generation-demand imbalance, voltage fluctuations, and frequency oscillations caused by the intermittence nature of these sources and lack of system inertia. This paper introduces a multifunctional two-stage grid-connected PV power system based on detailed mathematical modeling and innovative control techniques to provide efficient ancillary services to meet the grid code requirements and to overcome the issues of photovoltaic generation. The simulation results show that the designed PV power system is capable of operating at MPPT and suboptimal points to ensure the ancillary services of ramp-rate control, active and reactive power control, frequency regulation, and ac voltage regulation, under varying operating and atmospheric conditions.
Convolutional Neural Networks (CNNs) revolutionized computer vision and reached the state-of-the-art performance for image processing, object recognition, and video classification. However, the intensive processing nature of CNN hinders its adaptation in the resources limited edge devices. Artificial Neural Networks are known to be error tolerant, hence, tradeoff between accuracy, performance, power, and latency to meet target application is applicable. This paper proposes the Spatial Locality Input Data method for computational reuse during the inference stage for a pre-trained network. The method exploits input data spatial locality via skipping partial processing of the multiply-and-accumulate (MAC) operations for adjacent data. The computational data reuse was evaluated on three well-known distinctive CNN structures and datasets: LeNet, CIFAR-10, and AlexNet. The computational data reuse method saves up to 34.9%, 49.84%, and 31.5% of MAC operations while reducing the accuracy by 8%, 3.7%, and 5.0% for the three models mentioned earlier, respectively.
With the advances of space exploration vehicles that consists of payloads and subsystems, the most crucial one is the Electrical Power Subsystem (EPS) as it is responsible to provide power to the payloads and loads. As the scientific community focuses on improving the existing satellites technologies in order to make them suitable for future and longer missions, the most promising technique is to model comprehensively the electrical power system so it has a better reliability, efficiency and stability. This paper will present a full model of the EPS of MYSAT-1, the first earth observation CubeSat built at Khalifa University-YAHSAT lab, including the primary power source, energy storage, energy management and loads using MATLAB/Simulink platform. Furthermore, the simulation results of the load dynamics will verify the EPS design along with its power management system.
Silicon photonics is an emerging technology in electronics computing. It consumes comparatively less power and can perform data movement and computations at a much faster rate than the conventional electronics approach. Electrons can travel up to a speed of 1000 m/s in silicon while photons can travel 100 thousand times faster - an immense increase in the current state of the art. Because of this, in the last few decades, Silicon photonics is also thought of as an approach to accelerate AI (Artificial Intelligence) hardware which is conventionally time and energy consuming. Silicon photonics can perform the computation of AI algorithms at a lightning speed. In the current state of the art, ANN requires considerable MAC (multiply-accumulate) operations, and this could be done by using optical interferometers and interconnects on silicon chips with the present semiconductor fabrication process.
