Projects 2023

This project focuses on developing cutting-edge deep learning tools to determine whether a cancer (lymphoma) patient is transitioning to a more severe level of disease based entirely on demographic, laboratory, and medical imaging data. To date, a biopsy has generally been required to identify this transition, with patients often undergoing several biopsies over the years, resulting in complications and delays in treatment. As such, developing deep learning techniques to identify this transition using data acquired non-invasively (e.g. imaging, demographic, laboratory), would represent a significant step forward in patient care and AI-based personalized medicine.

The primary objective of this project is to develop a deep learning model to predict histologic transformation using demographic, laboratory and imaging data (alleviating the need for invasive biopsies). The student will work closely with graduate students in Prof. Arbel’s lab, a Research Scientist at MILA, and collaborating physicians at the Jewish General Hospital and the McGill University Health Centre.

This project focuses on developing cutting-edge deep learning tools for personalized medicine in multiple sclerosis. This involves developing deep learning techniques that can accurately predict patient-specific disease trajectories across multiple treatments. If translated into the clinic, the resulting tools would enable the most effective therapy to be identified quickly, and would represent a significant step forward in patient care and AI-based personalized medicine. Key to the success of the project is to develop modern generative modeling techniques to construct compact latent representations that effectively compress high dimensional medical images over time. These representations will then be used to predict patient-specific temporal trajectories, as well as to generate the associated Magnetic Resonance Images (MRI).

Students will work closely with graduate students in Prof. Arbel’s lab and a Research Scientist at MILA, and assist with (1) the development of generative modeling techniques (e.g. Latent Diffusion Models) with the aim of building deep learning representations for personalized medicine in multiple sclerosis, and (2) with supporting infrastructure.

Engineering identity is described as the process of ‘identifying with engineering, developing an engineer identity, and becoming an engineer’. Undergraduate students who had stronger identification were more likely to report that they would continue their education and professional career in engineering-related fields. Moreover, engineering identity has been established as an important consideration to retain females, minorities, and other under-represented students in engineering. There is emerging evidence that shows that different teaching and learning strategies are more effective at contributing to equitable learning outcomes. Specifically, educators can adopt more inclusive practices to counter systemic and structural barrier. Research has investigated how students of different demographics (e.g., gender, ethnicity, etc.) view their engineering identity. At the same time, studies have explored how the use of different teaching strategies impacts learning on students of different demographics. However, there has been no systematic study that investigates a link between equity and inclusivity in engineering education and engineering identity. The objective of this project is to establish a framework for investigating this link and in particular, to determine whether or not models used to predict how engineering identity impacts career aspirations/academic success of undergraduate students can be mediated by equity and inclusivity in engineering education.

The SURE student will participate in the following: 1. Complete a literature review on engineering identity, including the different tools (e.g., survey questionnaires) used to establish engineering identity 2. Complete a literature review on equity and inclusivity in undergraduate STEM education, with a specific focus on engineering education. Specifically, the study will identity the instructional strategies and practices that are more inclusive and that promote greater equity in terms of learning outcomes. 3. Assist in developing new tools (survey questionnaires) that can be used to measure engineering identity, student experience with equity and inclusion in engineering education, and student career aspirations and/or academic success.

Study of how shoppers move about in convenience stores, and what objects they pay attention to when searching for a specific product. The study will involve validating a neural network model of human visual attention and distraction during search tasks. Tools to be used include: - glasses mounted eye-trackers - person tracking software giving location and skeleton pose model of each shopper

- characterize data provided by glasses-mounted eye-tracker in the store environment - align eye-track data with person trajectory data and a 3D digital twin model of the store - run controlled eye-tracking experiments with select people doing search for a product in the store - compare observed behaviour with predictions made by neural network attention model

Internet Multimodal Access to Graphical Exploration (IMAGE) is a project aimed at making internet graphics accessible for people who are blind or partially sighted through rich audio and touch. We use rich audio (sonification) together with the sense of touch (haptics) to provide a faster and more nuanced experience of graphics on the web. For example, by using spatial audio, where the user experiences the sound moving around them through their headphones, information about the spatial relationships between various objects in the scene can be quickly conveyed without reading long descriptions. In addition, rather than only passive experiences of listening to audio, we allow the user to actively explore a photograph either by pointing to different portions and hearing about its content or nuance, or use a custom haptic device to literally feel aspects like texture or regions. This will permit interpretation of maps, drawings, diagrams, and photographs, in which the visual experience is replaced with multimodal sensory feedback, rendered in a manner that helps overcome access barriers. For the SURE Program, we are offering two positions to contribute to new avenues of research being pursued within the project in conjunction with our industrial partners. Depending on the experience and skills of the students we will hire, different research tasks and deliverables from the list below will be selected.

- design of audio-tactile experiences with a refreshable pin array or a kineshetic force-feedback device - design and development efforts to support IMAGE functionality on mobile devices - investigation and performance characterization of ML and computer vision modules - observation, analysis and design improvements to mitigate onboarding challenges of new users to the system - participation in user testing sessions to evaluate user experience under real-world use conditions

Based on ongoing research in our lab involving conversation with avatars, this project will develop a prototype platform to enable near-real-time conversations with historical personalities, using a combination of AI tools for avatar rendering and dialog management. For the SURE project, we are targeting primarily educational applications, helping to engage students with their learning by giving them the opportunity to, for example, discuss the principles of physics with Newton or work thorugh concepts of electromagnetism with Maxwell. The selected student will work on a subset of the research tasks and produce a subset of the deliverables listed below.

- determine how best to apply a generative image platform to produce suitably appearing historical figures as selected by the user - experiment with and select optimal configurations of a language model to provide educationally useful conversational interaction with the selected historical figure - select among available options a suitable avatar animation and voice synthesis framework

The Ai-Digital Nurse Avatar (ADiNA) project will develop the initial phase of avatar technology intended for interaction with seniors in nursing homes, home care, or retirement communities. The primary objective is to build AI-based tools that will provide assistance to nurses and other care staff, helping reduce workload by serving as a possible initial point of communication with clients, and triaging communications during periods of overload. The avatars, potentially presenting different on-screen human appearances and voices, as best-suited to the preferences of each client, will collect information through natural conversation and video-based interaction. The relevant information would then be conveyed to nursing staff in an appropriate format, without necessitating travel to every client for every interaction. A possible use-case scenario is that of answering a client in distress. This would be much like a nurse responding to a hospital “call button”, but with the potential of having the AI address non-critical issues, and otherwise, forwarding the calls to nursing staff by level of urgency. Another such scenario is to have the AI avatar carry out a subset of tasks required during regular staff rounds, serving as the nurse’s assistant by checking vitals and verifying the client’s general psychosocial state, to the degree that this is feasible through basic visual observation and dialog.

The student will work to support activities of a team of existing research personnel, focusing on actitivities related to one of the following tasks: 1. acquiring metrics of the senior’s wellbeing, and comparison against baseline models 2. extending the conversation management infrastructure (likely to be a third-party software architecture)

Various large scale datasets of molecular profiles of cancer cell lines (CCLs) and their response to hundreds of compounds and drugs have become available recently. In this project, we aim to develop deep learning (DL) approaches to predict and understand responses to compound/drug perturbations, by leveraging multiple such datasets. In particular, we are interested in integrating data from three large-scale datasets of CCLE, CTRP, and GDSC using domain adaptation and transfer learning approaches to improve drug response prediction performance.

- Conducting research - Implementing deep learning models - Data processing and cleaning - Analyzing the results and benchmarking against alternative methods - Writing a report

Novel therapeutic peptides are an exciting class of drug. They are short amino acid chains that have been shown to be effective for the treatment of various diseases. We’re interested in evaluating the utility of our method, RAPPPID (Regularised Automatic Prediction of Protein-Protein Interactions using Deep learning) for the purpose of developing such therapeutic peptides. To do so, we’ve created an online portal which dynamically serves the results of the RAPPPID model over the internet. The RAPPPID model, however, results in a lot of data that, in order to generate useful insights, must be cross-referenced with large public databases. Furthermore, in order to make conclusions about the data, we must visualise the data in a useful manner. Achieving this efficiently through a web-based platform poses its own unique challenges and constraints. Students who participate in this project will be responsible for conducting research into the effectiveness of RAPPPID for therapeutic peptides through the design and development of web-based data-visualisations of the results of our online platform. Students will be well-prepared for this research project if they have previous experience with visualising data, manipulating large datasets, and web-based technologies like client-side Javascript and charting libraries such similar to chart.js, plot.ly, D3.js, and the like. A basic understanding of or interest in molecular biology will be an asset to students who undertake this project.

- Investigate efficient, web-based data visualisations of predicted protein networks - Evaluate the effectiveness of RAPPPID for therapeutic peptide discovery using the designed data visualisations

To model the electromagnetic fields accurately and efficiently within sophisticated microstructures of modern engineering systems, devices, and biological structures high performance computing (HPC) methods, such as parallel and distributed simulations on emerging multicore/manycore platforms, are deemed promising for overcoming current computational bottlenecks. While robust and reliable 3-D automatic mesh generation procedures and solution strategies for electromagnetics are emerging, major computational challenges still remain for effective parallel and distributed 3-D adaptive finite element methods (AFEMs). Uniting AFEMs and HPC methods to achieve high gains in efficiency makes it possible to solve previously intractable problems; however, effective implementation of such techniques is still not well understood. AFEMs for parallel/distributed computing introduce complications that do not arise with simpler solution strategies. For example, adaptive algorithms utilize unstructured meshes that make the task of balancing processor computational load more difficult than with uniform structures. To help address these challenges, machine learning based approaches will be leveraged for developing state-of-the-art mesh generation for complex systems including biomedical applications.

The students in this project will research and develop efficient machine learning-based parallel and distributed adaptive algorithms for unstructured meshes that use complex data structures for implementing dynamic load balancing strategies for HPC environments such as multicore/manycore architectures. The students’ role will include involvement in all aspects of the engineering research process for this project including actual implementation of algorithms as executable code.

In this on-going research project, we consider how to design a broadband wireless access communications system that can adaptably adjust itself to the continuously changing complex environment by using machine learning (ML) techniques. We aim to explore the potential of applying ML techniques to harvest relevant environmental information for improving the resource allocation, performance and operation of the corresponding broadband wireless access communications system. Relevant environmental information can include weather, terrain, propagation (e.g., power, frequencies, etc.), social relationships (e.g., user groups, social networks, etc.) Various ML-based algorithms within a prototype testbed will be developed for the specific topics such as 3D channel modelling/estimation, hybrid ARQ, hybrid massive-MIMO precoding/beamforming, etc., to demonstrate the effectiveness of the Artificial Intelligence (AI)-augmented systems in terms of performance benchmarks such as energy consumption, increase in achievable capacity, reduction in interference, etc. Students will have a chance to understand various new concepts and development tools in both wireless communications (channel modelling, antenna array, beam forming, MIMO, etc.) and machine learning (deep neural network, reinforcement leaning, etc.), and to be involved in practical prototype development, and testing. As an example, one sub-project aims to develop a learning mechanism to jointly optimize Location & signal Power Allocation (PA) in UAV (Unmanned Aerial Vehicle) assisted MU-mMIMO communication system. Other sub-project aims to make use of both the terrain and weather information available in many sources such as Google map, online meteo, to develop a ML-based channel estimator for dynamic resource allocation in a broadband wireless access communication system.

Study the general concepts of ML and wireless communications. Learn how to search for and read scientific papers on a given signal processing or machine learning methods. Investigate Matlab toolboxes, PyTorch, Keras, Tensorflow, and DSP/FPGA hardware for possible applications to algorithm/prototype implementation. Assist in implementation and testing algorithms/prototypes, and in collecting, documenting and commenting the test results. The following skills and experiences are great assets: software development/testing, antenna design, Matlab, Python, VHDL, etc.

Full-Duplex Massive Multi-Input Multi-Output (FD-massive MIMO) Active Antennas Arrays (AAA) are considered for the next-generation broadband communications. Using a massive number of antenna elements can (i) help to adaptively create narrow beams continuously steered to follow the target user while avoiding interference from other users, (ii) increase the communications range, and system capacity. The smart AAA system can follow the mobile user based on (i) a hybrid 2-stage digital and RF precoding structure to reduce the complexity, and (ii) a full-duplex operation for simultaneous transmission/reception over a frequency slot to enhance both spectrum utilization and latency. In this on-going project, we investigate, design and test new promising antenna 3-dimentional structures (such as metamaterials, EBG, dielectric filled, etc.), with integrated power and low noise amplifiers, as well as RF combiners/splitters and smart DSP based control sub-system. The hardware testbed consists of powerful multi-FPGA, multi-microprocessor, and RF Analog-to-Digital Converter (ADC) and Digital-to-Analog Converter (DAC) modules to be programmed with digital signal processing algorithms for transmission and reception/detection of real wireless communications signals.

Full-Duplex Massive Multi-Input Multi-Output (FD-massive MIMO) Active Antennas Arrays (AAA) are considered for the next-generation broadband communications. Using a massive number of antenna elements can (i) help to adaptively create narrow beams continuously steered to follow the target user while avoiding interference from other users, (ii) increase the communications range, and system capacity. The smart AAA system can follow the mobile user based on (i) a hybrid 2-stage digital and RF precoding structure to reduce the complexity, and (ii) a full-duplex operation for simultaneous transmission/reception over a frequency slot to enhance both spectrum utilization and latency. In this on-going project, we investigate, design and test new promising antenna 3-dimentional structures (such as metamaterials, EBG, dielectric filled, etc.), with integrated power and low noise amplifiers, as well as RF combiners/splitters and smart DSP based control sub-system. The hardware testbed consists of powerful multi-FPGA, multi-microprocessor, and RF Analog-to-Digital Converter (ADC) and Digital-to-Analog Converter (DAC) modules to be programmed with digital signal processing algorithms for transmission and reception/detection of real wireless communications signals.

Multi-Input Multi-Output (MIMO) has been used in wireless systems such as LTE, WiMAX, Wi-Fi. Full–Duplex massive-MIMO (FD-massive MIMO) is considered for the next generation cellular communications (especially to support broadband Industrial IoT, M2M applications). In this on-going project, we measure and characterize MIMO Self-Interference channels in both microwave bands (for wider signal penetration application) and mmWave bands (for wider bandwidth, high peak data rates applications) and different practical scenarios in order to understand the implications on the FD-MIMO design requirements, in particular, on RF self-interference cancellation. We will investigate various types of channel environments: controlled free-space (e.g., in anechoic chamber), simulated rich scattering (e.g., reverberation chamber), or practical indoor and outdoor environments. Students will have a chance to understand MIMO systems, RF Beamforming based and Baseband Self-Interference Canceller design, Self-Interference and Intended Signal channel measurements, and to work with real-life measurement facilities and testbeds.

Study the general concept of massive-MIMO, radio-wave propagation in free-space and in rich-scattering environments; learn the operation principle of measurement equipment/facilities such as vector network analyzer, spectrum analyzer, vector signal generators, anechoic chamber, reverberation chamber and/or MIMO testbeds; prepare measurement set-ups; assist graduate students and/or research associates to evaluate/analyze measurement/simulation results. Learn how to use Matlab and SystemVue simulation tools.

For power-efficient operation, RF Power Amplifiers (PA) should operate near the saturation region, but this creates non-linear behaviors and distortions in amplified complex signals. Typical approach to balance power efficiency and performance degradation is to use PA linearizers (or PA DPD, PA Digital Pre-Distorters). Non-linear power amplifier with deep memory and very wide operating bandwidth, such as in GaN new power efficient PA technology, is challenging to linearize using conventional models and DPD techniques. In this on-going project we will continue to investigate/test existing state-of-the art practical FPGA/DSP based PA DPD techniques with a single and a network of GaN and GaAs PAs. After the established performance reference, we will develop, simulate and test Deep Neural Network (DNN)-based algorithms to improve and simplify simultaneous linearization of multiple PAs in Active Antenna Arrays (AAA) of massive-MIMO systems.

Study about Power Amplifier characteristics, characterization, measurements and Subsequent Modeling in Matlab. Learn about typical PA parameters such as: gain compression (P1dB compression point), Amplifier Saturation Output Power (P3dB), IMD, IIP3, OIP3 (input/output third-order intercept points), AM/AM and AM/PM distortion. And also about other concepts such as: ACPR (for modulated signals, like QPSK or QAM) and Error vector magnitude (EVM) (for modulated signals, like QPSK or QAM). Help with configuration and interfacing of DPD platforms with a multi-PA AAA m-MIMO testbed, as well as with running the experiments, and pre-analyzing the results under supervision for graduate students. Review literature other conventional and new ML based PA linearization techniques.

The SURE project uses applied machine learning to design next-generation photonic integrated circuits. Photonics brings the speed of light to applications such as high-performance quantum and neuromorphic computing, self-driving vehicles, biomedical sensors, and high-capacity data centers. Designing compact, robust photonic circuits by hand is non-intuitive; therefore, a big opportunity exists for machine learning to make sense of complex patterns in data and become an integral tool in the design cycle.

The qualified intern will join our team in creating a “virtual” nanofabrication environment using machine learning. Deep neural networks serve as a high-resolution virtual nanofabrication facility that allows prediction of the fabrication outcomes and pre-emptive correction of anticipated deviations—resulting in previously unfound levels of performance and massive savings in time and cost for future designs. The role of the intern will be to help improve the performance and capabilities of this new methodology. They will have a fundamental understanding of electronics/photonics, nanofabrication, programming (e.g., Python, JavaScript), and/or machine learning.

The fabrication of ultra-short channel transistors, with channel length ~10 nm and below, is challenging for conventional lithographic methods. By clever use of self-assembled monolayers (SAMs), lithographic resolution of ~10 nm can be achieved. In this project, facile SAM based lithography and ultra-thin metal oxide deposition methods will be used to develop short channel monolayer and bilayer graphene field effect transistors (FETs) and ion-sensitive field effect transistors (ISFETs).

The student will work with a graduate student in the research group to fabricate ultra-short channel monolayer and bilayer graphene FETs. Research tasks include: 1) the test, characterization and optimization of nanofabrication protocols, including SAM deposition, metal deposition, graphene deposition, and metal oxidation 2) measurement and analysis of fabricated devices using semiconductor parameter analysis and impedance spectroscopy The research tasks will entail experimental research work within Prof. Szkopek's laboratory and shared access user facilities at McGill University.

The charge-trap transistor (CTT) is a non-volatile memory device that can be used to store analog memory. In the context of neuromorphic (brain-resembling) systems, the CTTs store synaptic weights and support in-memory computation. They rely on charge trapping within the high-k dielectric gate oxide to adjust the threshold voltage of the device in a non-volatile manner. Preserving the adjusted value of the threshold voltage of CTT devices is a critical challenge for realizing high accuracy storage of synaptic weights. It is, therefore, of the utmost importance to be able to verify that the threshold voltage of individual CTTs is adjusted to the correct value, refresh the CTT threshold voltage level (to counter leakage), and control the level of the threshold voltage of the CTTs. There exist several potential integrated circuit solutions which can be designed to provide program/erase control and verification for the threshold voltages of individual CTTs. The objective of this research project is to design circuits to enable peripheral adjust-verify and control CTT threshold voltage drift. This circuitry must be integrated with an array of CTTs and should be able to support programming and erase, verify adjusted weights, and correct threshold voltage drifts for individual CTTs and larger arrangements of CTTs. Furthermore, this weights adjustment circuitry will be integrated with the rest of the neuron circuits, previously designed by our team.

The student will perform literature review on the CTT. Design and implement the circuits to support weight adjustment of a CTT array using Cadence Virtuoso in GlobalFoundries 22nm technology. Student will simulate the design under parameter variation and verify functionality and performance. Student will produce paper/poster to summarize results and conclusions.

Existing hardware designs of leaky-integrate and fire (LIF) neurons are area-hungry and thus limit scalability, adaptability, and efficiency. To enable scalable neural networks in hardware, a novel neuromorphic architecture based on compute-in-memory devices is proposed. This mixed-signal architecture should include an analog weight storage array and a digital neuron. The objective of this research project is to design a digital circuit of an LIF neuron, which will then be integrated within the neuromorphic architecture. The circuit should be able to keep track of the accumulated charge from the input signals and trigger an output pulse once a digital threshold has been met. The circuit should also be able to continuously leak and reset itself once the output pulse has been generated.

The student will perform literature review of the neuromorphic architecture. The student will design and implement the digital circuits of the LIF neuron using Cadence Virtuoso in GlobalFoundries 22nm technology. Student will simulate the design under parameter variation and verify functionality and performance. Student will produce paper/poster to summarize results and conclusions.

The inherent advantages of three-dimensional (3D) integrated circuits (ICs) are well-aligned with the continuous demand for increased density of functionality, reduced latency and power dissipation of communication, and heterogeneity of modern applications. Delivering power efficiently to highly heterogeneous voltage domains across the tiers of a 3D IC is, however, a significant challenge. Integrated power delivery methodology is a promising approach for realizing an efficient and robust power delivery system 3D ICs. High-density passives (inductive and capacitive) are critical components of an integrated power delivery system. This research project has two main objectives: 1) surveying the state-of-the-art integrated inductive and capacitive passive technologies and 2) implementing novel integrated passives compatible with integrated power delivery in 3D ICs. The key features of the targeted integrated passives include high density (small footprint), orientation with standard processes for the fabrication of 3D ICs, low parasitics, supporting broad power and frequency range, and high quality factor.

(1) A literature review of the state-of-the-art in integrated inductive and capacitive passive technologies, including a summary of their numerical/parametric models, performance metrics, advantages, and disadvantages (2) Design and simulation of integrated passives using finite element method (FEM) tools, such as Ansys Maxwell, HFSS, and Q3D (3) Extraction of numerical/parametric expressions of the designed passives (4) Documentation of deliverables and poster

The gaming market is emerging. The global gaming market size was worth 202.64 billion in 2021 and is expected to expand continuously. Quality issues in games can induce bad user experience and affect game profits. For example, the first release of Cyberpunk 2077 had significant quality issues (e.g., blocking issues). The company refunded users who were not satisfied with the game. In this project, the student is going to conduct a survey on the quality assurance practices of games. The student will collect a set of existing papers on related topics and a set of open-source games. The student will analyze the collected materials to analyze and characterize the quality assurance practices of game software.

1. Collect papers related to game quality assurance 2. Collect open-source game subjects 3. Review the collected papers 4. Analyze the quality assurance practices (e.g., testing, code review, etc.) in the open-source game subjects

An emerging number of Android apps are relying on cloud services as a backend. Cloud providers are providing diverse services for Android apps (e.g., cognitive AI APIs, Cloud storage, etc.). To authenticate communications between apps and cloud services, app IDs and app secrets are often used as the sensitive identifiers of the Android apps. Such sensitive identifiers (or cloud service secrets) should be kept confident. However, a large number of app developers failed to properly protect their cloud service secrets. They embed such secrets in their app code without encryption, making them easily leaked via simple reverse engineering. In this project, the student is going to analyze the leaked cloud service secrets in Android apps and develop a technique to detect such leaked secrets. The student is expected to produce a list of popularly used cloud services in Android apps and focus on detecting leaked secrets of the identified cloud services. In this project, the student will learn program analysis techniques and gain hands-on program analysis experience from developing the cloud service secret detector.

1. Study the usage of cloud services in Android apps 2. Identify popular cloud services used in Android apps 3. Develop a cloud service secret detector for Android apps