Innovative Projects Realized

AI based technology adoption in circular economics

This project is a cross-disciplinary study of econometrics and machine learning (ML) models applied to the decision making modelling in industry. The problematic arises from the lack of tools supporting the transition to circular economics model and the need to identify the key factors to influence this transition.

The project aims to explore the key elements affecting the high management decision making process in technology adoption. A discrete choice experiment survey will be conducted. The data analysis procedure will involve both econometrics and machine learning techniques. Canadian partners will provide the information sources in industrial domain, as well as the knowledge in advanced ML and AI modelling techniques. French side posesses all the required knowledge and skills for analysis of human behaviour. The simulation and theory-testing framework proposed in previous works, will constitute the core of modelling approach, allowing to increase the reliability of results.

Exploring the boundaries of serial electron diffraction

Serial electron diffraction (serial-ED) crystallography is an emerging structural biology method whereby data are collected from protein nanocrystals using a sub-micron-sized electron beam. The method is set to lead the future of nanocrystallography, given its cost-effectiveness and the small amount of sample required. Combining the expertise of the Colletier and Miller teams, the project will address three issues of fundamental importance to further advance the serial-ED methodology. Specifically, we will (i) test if the method can be applied to solve nano-crystalline protein structures in the cellular environment (i.e. in vivo); (ii) determine what is the smallest crystal size that can be probed by serial-ED; and (iii) offer a proof of feasibility for time-resolved serial-ED experiments on macromolecular nanocrystals. Results from our Serial-ED-boundaries project will allow to explore the boundaries of serial-ED and to develop its full promise for nano-crystallography.

Open-air fabrication of dynamic cantilever gas sensor

Sensitive gas sensors are urgently needed for the detection of biomarkers for disease prevention and greenhouse gas emissions for environmental monitoring. Dynamic cantilever gas sensors work by oscillating the cantilever to its natural resonance frequency by typical MEMS (micro electromechanical system)-based actuation methods. The adsorption of a target gas on the cantilever increases its mass and shifts the resonance frequency. Conventional cantilever-type sensors are comprised of a silicon structural layer with a thickness on the order of 1-10 microns. The silicon cantilever is then coated with a thin layer of a receptor material to enable adsorption of specific gases. The project involves the open–air fabrication of innovative, highly-sensitive cantilever gas sensors by Atmospheric Pressure Spatial Atomic Layer Deposition (AP-SALD). Different oxide materials will be deposited by AP-SALD and used to fabricate cantilever gas sensors, where the oxide material acts as both the structural layer and the gas-sensitive receptor layer. The cantilevers will be characterized and tested for sensing of biomarkers and methane, a potent greenhouse gas.

Making AI Ready for Safety-Critical Applications

This project is a collaborative endeavor of researchers (6 supervisors, 3 PhD students and one postdoctoral fellow) which will be either members or visitors of the incoming “International Laboratory on Learning Systems” (ILLS) of the CNRS (starting in early 2022) with Université Paris-Saclay, McGill University and École de technologie supérieure (ETS). We will develop rigorous techniques for building safe and trustworthy AI systems, establishing confidence in their behavior and limitations, thereby facilitating their successful adoption in society. Our research axes will address three fundamental problems: How to detect errors in image segmentation algorithms? How to learn more with less data? How to quantify the information leakage of trained software? We are expected to devise new methods for preventing errors in autonomous cars and medical imaging systems; learning from little information; and auditing privacy risks of AI algorithms. Research activities will be constructed to facilitate the design of truly interdisciplinary research between Computer Sciences, Electrical Engineering and Applied Mathematics, where the potential for innovation is greatest. Our results will be disseminated at the flagship conferences and in prestigious journals of machine learning.

Restoring bacterial sensitivity to aminoglycosides antibiotics

Bacterial resistance to aminoglycoside antibiotics is a major public health problem. The WHO (World Health organization) has highlighted the urgent need to find new treatments to fight multidrug resistance MDR bacteria, as actual drugs are less effective or inactive. This research project aims to develop new chemical compounds to restore bacterial sensitivity to antibiotics currently used in clinic. Preliminary results have shown that small designed molecules are able to restore the sensitivity of resistant bacteria to aminoglycosides. Moreover, the efficacy of these compounds to restore sensitivity to aminoglycosides of resistant clinical bacteria (such as pathogens involved in serious illnesses: Cystic Fibrosis, pyogenic liver abscess, bloodstream infections or urinary tract infections) will be investigated as well as the potential toxicity for human cells in order to prevent possible side effects.

Attack Detection for 5G Networks using AI/ML

The roadmap for 5G networks is already taking shape due to several industrial and academic research efforts. 5G networks are expected to support more diversified services, which should create exciting business opportunities in many vertical sectors. Achieving this requires improving the technologies behind the evolution of 5G and leveraging machine learning (ML)/artificial intelligence (AI) techniques for efficient management of network resources. However, to meet the requirements of various industries, 5G should not only rely on new enabling technologies but also on providing a secure network architecture beyond current network designs. The massive explosion of connected devices which formed what is known as the Internet of Things (IoT) makes this requirement more challenging. To overcome this issue, 5G architecture needs to detect attacks and anomalies before they occur more intelligently. Traditional anomaly detection mechanisms may not scale with the 5G needs. Hence, in this research project, we are motivated to apply machine learning/artificial intelligence techniques to detect attacks and anomalies and thus contribute to designing a secure 5G network. Combining security with artificial intelligence is indispensable nowadays. Indeed, the benefit of the project would be to design and implement innovative and novel techniques for intelligent anomaly detection for 5G networks.

The effects of AI empathic music therapy on psychosocial and cognitive functions of older adults: a pilot and validation study

Music therapy is an evidence-based effective intervention to promote psychological, social and emotional health, and it is specifically favored by older adults. However, little is known on the efficacy of the AI-assisted empathic music therapy in which the music is continuously adapted to individuals’ mood. To fill this gap, the proposed project will assess the added benefits of the AI music therapy (AIT) compared to traditional music therapy (TMT) in older adults. It is a 3-arm randomized controlled trial (RCT) in which 45 eligible participants will be evenly randomized into the AIT, TMT, or a no-intervention control group (CG). All groups will complete outcome assessments at pretest, posttest, and a 3-month follow-up session. The two therapy groups receive a 2-week music therapy (4 sessions per week, 30 min per session) between the pretest and posttest. The outcome measure data will be analyzed to assess the training benefit and its durability.

Carbon coating to improve Titanium and Aluminum powder properties for LPBF application

Aluminum and titanium components fabricated via laser powder bed fusion (LPBF) process have gained industry interest, especially for automotive and aerospace applications, due to their enhanced lightweight and mechanical properties. For a higher part quality and surface resolution, the use of fine powders is desired for the LPBF process. However, the poor flowability of fine powders produces a major processibility challenge. To improve the powder properties, the dry coating technique is suggested to carburize the aluminum and titanium alloys. Hence, the objective of this project is to optimize the coating parameters and powder processibility for the LPBF process. The proposed methodology starts by testing the coated powders for powder flowability, packability, coating thickness, and compositions to down select the optimum coating conditions. Then, the selected and virgin powders will be used to print the test specimen for the physical and mechanical tests to evaluate the powder processibility for the LPBF process.

Feasibility of individualised synthetic speech for children with complex communication needs (CCN) in three South African languages

An individual’s voice is unique due to a combination of their physical and social characteristics. For children who present with complex communication needs (CCN), sometimes the only functional way to communicate is by using an augmentative and alternative communication (AAC) device. AAC devices with speech output capability are known as voice output communication aids (VOCAs). The use of VOCAs have been shown to increase quality of life for individuals with speech impairments. However, many users lack a personal connection with the voice on their device. In addition, the languages offered are not reflective of South Africa’s (SA) rich variety of official languages. The aim of this study is to develop a comprehensive and feasible method for producing the most intelligible and natural-sounding individualised synthetic voices for SA children with CCN who speak South African English, Afrikaans and isiXhosa, by using existing open-source speech synthesis software. The Mitacs Globalink internships falls part of a larger PhD study. There are four research objectives in the PhD. However, the internship will concentrate on the third objective: Describe the effects and modifications of an open-source speech synthesis system, when developing tailor-made, age-appropriate synthetic voices for children in three languages.

Evolutionary and functional analysis of genes involved in nitrogen-fixing symbiosis as a case of convergence

A major challenge facing farmers is obtaining the nitrogen needed to support plant growth. Inoculation of legume crops with bacteria known as rhizobia, which supply plants with the required nitrogen, is a green alternative to environmentally hazardous nitrogen-fertilizers. The development of highly-efficient rhizobium inoculants is a pre-requisite for sustainable intensification of agriculture. In five clades of legume, plants independently evolved an exploitative strategy associated with a high return on investment to the plant and is thus an ideal pathway to manipulate to optimize the symbiosis. This exploitative strategy is based on plant-produced antimicrobial peptides and a rhizobial peptide transporter. The mechanisms driving differentiation in the other clades have not yet been analyzed. In this project, we will test the hypothesis that repeated coevolution took place between the different legume clades and their associated rhizobia, at both the phenotypic and molecular levels. In doing so, we will generate new knowledge to engineer a more efficient symbiosis for agricultural benefit.

Novel rhamnolipids for a sustainable bioeconomy

Today’s society faces several challenges in implementing and enforcing environmental protection for future generations. In this respect, supplementation of petro-based chemicals by bio-based ones is of special importance to reduce the consumption of crude oil. Among such measures, chemical surfactants can be replaced by biosurfactants, e.g. rhamnolipids which are produced via fermentation from renewable resources and used as ingredients in household detergents, pharmaceuticals, or cosmetics. Thus, approaches to increase the diversity and availability of tailor-made biosurfactants will promote the transition to a bio-based economy. In this project, we will investigate potentially O-methylated rhamnolipids which are less aggressive than their chemical counterparts and anionic “conventional”” rhamnolipids. Specific objectives are the production and structural determination of O-methylated rhamnolipids from unexplored bacteria of Burkholderia species and the introduction of the respective biosynthetic genes into a non-pathogenic bacterial host to establish a production platform. The results obtained will help to achieve a sustainable economy and will further boost the existing research collaborations between Canada and Germany.
“

Artificial Intelligenece to study tumor heterogeneity

High throughput multi-omic cancer studies have described the inter-tumor heterogeneity and led to well defined molecular classifications. Nevertheless, these classifications only reflect the most abundant tumor subtype in the examined sample, thus neglecting intra-tumor heterogeneity, a major source of therapeutic resistance. As advanced microdissection techniques to isolate a cell population of interest from heterogeneous clinical tissue are not feasible in daily practice, bioinformatic tools to estimate intra-tumor heterogeneity are therefore urgently needed.

Our aim is to develop integrative models of tumor heterogeneity and to infer biological behavior and associated clinical indicators using new computational methods based on Artificial Intelligence. We propose to take up this challenge by developing original methods applied to the study of heterogeneous tumor samples using benchmark datasets. First, we intend to develop an approach inspired by machine learning methods to address the problem of subtype classification accounting, for intra-tumor heterogeneity. We will then assess the impact of multi-omic data integration and feature selection in tumor heterogeneity quantification. In parallel, we will study the spatial tumor heterogeneity using microdissections of complex tumor tissues. Finally, we will apply the developed algorithms to a large pancreatic tumor cohort and decipher the clinical impact of intra-heterogeneity and its spatial properties.