Self-Adaptive Pattern Recognition with Deep Neural Networks

The purpose of this project is to investigate self-adaptive forecasting and anomaly prediction algorithms based on deep neural networks (DNNs). DNNs present a compelling technology due to their wide-spread availability through open-source projects (e.g. TensorFlow, MXNet). However, usability of DNNs in scenarios outside of image, speech or text pattern recognition is mostly unproven. This project aims to reduce the knowledge gap that exists in the usage of DNNs in the context of pattern recognition with DNNs in network management and network equipment manufacturing.

Optimization of a calibration procedure for Mecademic’s Meca500 robot arm

Mecademic manufactures the smallest and most precise six-axis robot arm. The repeatability of this robot is better than 0.005 mm, but like any industrial robot, the robot’s accuracy is far worse. The only practical way of improving the robot’s accuracy is to calibrate each individual robot.

Pure-sine GaN-based motor inverter

This project applies wide-bandgap (WBG) transistors to voltage level multiplier module (VLMM) topology in motor inverter applications. It is expected that this approach can yield the benefits of WBG motor inverters (high motor efficiency, fast control response, lower motor torque ripple, close to ideal sinusoidal motor current waveform, smaller filter size, lower cost filter, etc.) while leveraging the benefits of VLMM (lower component cost, high frequency switching only at low voltage, filter-less output signal) to yield a commercially viable highly-efficient pure-sine motor inverter.

Wideband envelope detector for dynamic supply control in RFIC PA’s

Radio frequency integrated circuit power amplifiers (RFIC PAs) operating at microwave frequencies (e.g. 5 to 6 GHz) and at millimeter-wave (mm-wave) frequencies (e.g. 60 GHz) are electronic components used in the front-end modules (close to the antenna) of mobile communication equipment such as cellular handsets. Envelope detectors constitute a critical component in a newly proposed dynamic biasing technique for RFIC PAs based on positive envelope feedback, for power efficiency improvement and distortion reduction.

Efficient face recognition for wearable camera devices

Titan Sécurité Inc. has deployed wearable video camera devices for security and surveillance applications, and seeks to accurately detect and recognize objects appearing in captured videos. This project focuses on video-based face recognition (FR), where facial trajectories captured with video cameras are compare against one (or few) reference stills for each individual of interest. The performance of these FR systems is typically poor due to complex and changing video surveillance environments, e.g., variations of facial appearance due to pose, illumination, blur, etc.

Automating financial reports redaction

The objective of the proposed research project is to automate the redaction of financial portfolios reports. The generated reports should inform the reader about which factors influenced the portfolio’s returns, to what degree, and how far these factors deviate from the norm.

Energy Efficient 360 Video Processing for Portable VR-Technologies

This project focuses on the energy consumption of modern virtual reality applications. The target devices are virtual-reality-glasses that can be worn on the head and that simulate a virtual 3D environment to users. Most modern devices are still rather heavy, uncomfortable to wear, and attached to a powerline such that the user experience can still be enhanced. In this project, the goal is to make the glasses require less power and energy during operation such that the operating time and battery requirements can be minimized.

Natural Language Generation for Intelligent Tutoring Systems

This project solves the problem of generating content for a conversational intelligent tutoring system (ITS). The ITS gives questions to the student and then analyzes their answer using machine learning algorithms. Based on the student answer, the system will give them the hints and teach them how to solve the question. Moreover, the system will interact with teacher to generate content included questions, hints, and answers.

Phased array inspection of large size forged steel blocks

The objective of this project is to integrate ultrasonic imaging into nondestructive inspection process performed in industrial settings on large size steel blocks. Indeed, Fink Steel is specialised in manufacturing large size forged steel and need to warrant the quality of each part before shipping using ultrasounds. This imaging method based on the same principle as echography is commonly used for numerous industrial applications. However, the large dimensions involved in this project require new transducer design.

Exploring Deep Learning Architectures for Automatic Casting from Movies

Automatic casting applications aim is to accurately recognise facial regions that correspond to a same actor appearing in a movie to produce described video. This project will focus on the tasks of re-identify the face of each principal actors when they appear in different scenes of a movie. This is a challenging task because although recent movies are typically high resolution, the faces are often occluded and their appearance varies significantly according to pose, scale, illumination, blur, etc.

Pages