One of the main challenges in this domain is the detection and mitigation of nuclear threats, especially in big cities. Today, nuclear sources are not just restricted to the already well-controlled power plants or military facilities, but can be widely found in hospitals via medical equipment, medicine, and medical waste, or in research laboratories through research materials, or many other such sources. These sources when obtained distributedly or taken from a single source can easily be engineered to produce a radiological dispersal device or the “dirty bomb.” Moreover, these sources may be mobile, i.e., carried by individuals, and can be moved around through logistical chains, which poses a real threat to the population at large.

To address these challenges, we propose for the first time a multi-faceted approach to develop a well-rounded dynamic participatory system, using Internet of Things (IoT) and Mobile Crowdsensing (MCS), which can provide continuous high-precision and real-time detection, tracking, and localization of small mobile and immobile radiation sources and unexpected radioactive materials, within a noisy radioactive environment in an urban setting. The proposed solution also has the potential to cross the boundary of nuclear detection, as other sensing nodes and networks may be developed, on similar principles, for applications in environmental, industrial, or military monitoring.

The post Detecting, Localizing, and Tracking Radiation Sources Using IoT and Crowd-based Sensing appeared first on Khalifa University.

Most of the MRI-based computer-aided diagnosis systems proposed so far, indicate the presence of CaP either on whole MRI scan or in a sectional part of it without providing any information about the location of the CaP lesion in the MRI scan image. In this project, we propose to fill this gap by designing an advanced machine learning system capable of accurately locating the related lesions in the MRI scans. This information will provide pertinent guidance for the physician when performing the biopsy, thus reducing the risk of misdiagnosis.

The post Advanced Machine Learning Systems for the Early Detection of Prostate Cancer in Multi-modal MRI Images appeared first on Khalifa University.

Deep learning methods are a class of neural networks that allows a machine to be fed with raw data and to automatically discover the necessary formulas needed for classification. After training, the performance of the system is tested on new data sets. Deep neural networks are composed of a multi-layer stack of simple neural networks. Each module in this stack transforms the input using non-linear or linear mathematical operations. Despite the progress enabled by deep-learning methods, limitations of the technology remain to be explored and addressed in order for it to reach its full potential and range of application areas. In this project, we will design and develop novel systems that extends the capabilities of current technology.

The post Efficient Image Classification Using Deep Neural Networks and Sparsity-inducing Transforms appeared first on Khalifa University.

Majority of the EVs use permanent magnet motors as prime movers in the drive train of the EVs. Permanent magnets are compact in size and efficient but expensive. Permanent magnets are made up of rare earth materials and these are available in limited quantity and as the number of vehicles increase the cost of these magnets will rise exponentially. Therefore, there is a need to develop an alternate drive technology based on other motors like induction motors or switched reluctance motors.
This research proposal aims at developing a cost-effective drive train based on induction motors. Proper estimation of rotor and stator resistance and flux estimation, developing compact and efficient power converters, distributed drive train using in wheel type drive, better utilization of the battery voltage using dual fed drive and increased drive range with efficient regenerative breaking are some of the objectives of the proposed research proposal.

A challenge in using induction motor is the high DC bus voltage requirement, therefore a suitable high gain DC to DC converter to interface the low voltage battery bank to the high voltage DC bus of the inverter will be developed. This DC-DC converter need to be bidirectional for regenerative breaking. A mathematical model of the drive technology will be developed and verified through computer simulation. A scaled down model of the EV drive train will be designed to verify the proposed improvement experimentally.

The power electronics research laboratory at Khalifa University has the necessary infrastructure and facility to execute the above project. The research team has the necessary background and expertise, and is confident on the successful completion of the project in the scheduled time frame. The research team will consist of graduate and undergraduate student researchers (with preference to UAE nationals) and will be led by a team of three experienced faculty members, two of which are UAE nationals. The successful completion of this project will prove the feasibility of using induction motor as the prime mover in the drive train of EVs.

The post An Improved Regerative Drive Train for Electric Vehicles appeared first on Khalifa University.

The post Incentive Mechanism and Game Theoretical Model for Crowd Sensing Platforms appeared first on Khalifa University.

In many cases, it turns out that the obtained convex robust optimization problem can be cast as a second-order cone program (SOCP), or more generally, as a semidefinite program (SDP). Most (if not all) previous work essentially provides reductions from the underlying machine learning problems under uncertainty to robust convex optimization, but stop at this point since one can then rely on available general convex programming (CP) solvers to tackle the problem. However, it is observed that the computational efficiency of available general CP solvers is not practical for very large data sets. On the other hand, most of the problems arising in machine learning have a special structure that can be exploited to provide faster algorithms. This motivates the exploration of specialized CP methods, such as first-order methods. Such techniques have been extensively explored for the non-robust optimization versions of machine learning algorithms, but we are not aware of any extensions to the robust versions.

The first part of this project aims at delivering practical algorithms for such robust versions of common machine learning optimization algorithms, as well as theoretical and empirical analyses of their performance on real-world data. The situation becomes even more challenging when the solution space is discrete, e.g., in the case of clustering problems, in which case the problem is computationally hard. In the non-robust case, a standard approach is to formulate and solve a linear programming (LP) or SDP-relaxation of the problem, then use careful rounding techniques to map the obtained continuous solution into the discrete solution space without losing much in the objective value. Extending these techniques to the robust version makes the second main part of this project. Based on the obtained results, a software tool will be built to provide efficient algorithms for robust versions of a wide range of machine learning algorithms.

The post Efficient Algorithms for Robust Optimization with Applications to Machine Learning appeared first on Khalifa University.

In this project, we will take a cue from the authors’ most recent works to build and demonstrate a unified modeling framework capable of modeling soft and rigid robots within the same formulation, creating a standard for the modeling and control of soft robots. A complete and systematic generalization of the main concepts of robotics will be proposed. This unprecedented attempt will be based on a new modeling approach called Piecewise Constant Strain, which is a generalization of the geometric theory of robotics developed since Brockett’s original work on the subject. The result of this unification and generalization will be transferred into existing robotics simulation platforms in the form of toolbox and software packages. In particular, a MATLAB toolbox and SOFA packages will be developed for mechanical design and low-level control and high-level control, respectively.

The post A Standard Framework for Soft Robotics Modeling and Control: Filling the Gap with Traditional Robotics appeared first on Khalifa University.

Most of traditional architecture solutions of negative impact of memory access on performance and energy are focusing on employing multi‐level memory system and architecture. The ideal solution is to decrease the distance between the system memory and processing to zero. Memristor technology is an emerging Resistive Random Access Memory (RRAM) technology that holds great potential to play a role in achieving close to ideal solution by enabling In‐Memory‐ Computing (IMC) for both digital and analog type operations. In addition to reducing the memory bottleneck, IMC makes it possible for IoT type of systems to do local processing and decision making that enables autonomous application. The objective of the proposed research is to explore memristor-based architectures to perform machine learning for pattern recognition applications in IoT devices.

The post RRAM-based Computational Intelligence Hardware for IoT appeared first on Khalifa University.