Projets novateurs réalisés

Bridging the Gap Between CPUs and GPU with Out-of-Order SIMT

Most of today’s computers, from cell phones to supercomputers, are heterogeneous: they integrate processors that are optimized to quickly execute a few tasks (CPU
cores), and processors that can perform many independent tasks in parallel (GPU cores). GPU cores and CPU cores have different instruction sets: they understand
different languages. A task written for CPUs cannot run on GPUs, and vice versa. As a result, programming current heterogeneous architectures is challenging and few
applications can take advantage of the processing power offered by GPU cores. We will address this incompatibility by designing a hybrid CPU and GPU core, which presents the same instruction set as multi-core CPUs while offering the same parallel performance as GPUs.
To experiment and validate our proposition, we will model this new hardware in a software simulator mimicking the behaviour and timings of the proposed hybrid
architecture. We will develop the design using an existing CPU simulator.
This project will significantly ease the development effort required for designing applications which run efficiently on modern computers. It will also harness the
underutilized processing power available in contemporary hardware.

Découpage virtuel de corps viscoélastiques déformables à l’aide de la méthode SPH avec support GPU pour les logiciels de simulation chirurgicale

La découpe virtuelle de corps mous viscoélastiques (tel qu’un organe du corps humain) en temps réel constitue un problème d’envergure dans les logiciels de simulation chirurgicale. Le projet consiste à développer un algorithme de découpe 3D en temps réel offrant un rendu visuel réaliste à l’utilisateur et un retour de force optimal d’un dispositif haptique simulant l’outil d’un chirurgien. Deux méthodes de calcul des modèles viscoélastiques seront analysées durant ce projet : la méthode des éléments finis appliquée à un maillage constitué de tétraèdres et la méthode « Smoothed particle hydrodynamics » (SPH) utilisée généralement dans la simulation
de fluide. Ces deux méthodes seront accélérées à l’aide de processeurs graphiques (GPU). Nous souhaitons augmenter le réalisme du simulateur en offrant une immersion virtuelle optimale au chirurgien.

Fluorescently-labeled nanoparticle target for imaging

There is currently a great need to develop better imaging probes for studying biological process. In this proposed internship, Drs. Pacheco and Hashir will be evaluated an aluminum coated alloyed quantum dot system. These quantum dots are brighter than organic fluorophores and with the aluminum surface coating, these quantum dots are less pervious to environmental factors. This internship is beneficial to Market Link Scientific because they have a whole animal optical system and are evaluating possible applications of this instrument. The quantum dots are useful contrast agents for them and hence, the interns will assess the utility of these quantum dots probes in imaging applications.

Online Learning of Gait Models for Fall Prediction

Meaningful parameters can be extracted from data that describes a person walking as time progresses. Gait asymmetry is one such parameter that has been shown to be correlated to the likelihood of a person falling. More generally, the variability in gait parameters can be used for human fall prediction. Currently, models used to describe gait cycles do not generalize well to new variable data. The proposed method learns a model of the gait cycle during online measurement, using a rich and continuous representation that can adapt to inter and intrapersonal variability by creating an individualized model. We expect this model will estimate more accurate gait parameters for improved fall prediction. Furthermore, we predict the evolution of patient models could be used for early detection and assessment of other health conditions.

Small Molecule Ice Recrystallization Inhibitors as Cryo-additives for Red Blood Cell Cryopreservation

The transfusion of red blood cells (RBCs) is a lifesaving procedure for many patients. While the Canadian blood supply is safe and consistently meets demand, the primary method for the storage of RBCs prior to transfusion is refrigeration at 4°C for a maximum of 42 days, a process which can lead to inefficiencies and discards. Cryopreservation is a desirable method for the long-term storage of RBCs, providing access to a large quantity of RBCs required in emergency transfusions or for patients with unique transfusion requirements. However, current cryopreservation methods employ high concentrations of glycerol which must be removed prior to transfusion. Not only is the deglycerolization process costly and time-consuming, deglycerolized RBC units have a limited shelf-life of 24 hours. The proposed project will investigate the use of ice recrystallization inhibitors to optimize the cryopreservation of RBCs and improve access to frozen RBCs for patients in need.

Estimating reservoir changes in a heavy oil reservoir through application of anelasticity and rock physics analysis to time-lapse seismic data

Oil-sands reservoirs offer huge resource potential with very low decline rates compared to other unconventional exploration targets. Unfortunately, this comes with the trade-off of requiring high initial investment as well as high operating costs. In order for such projects to be economic in the long-term, it is essential that we monitor changes in the reservoir to maximize production and by extension, return.
Time-lapse (4D) seismic monitoring is a common, cost-effective means of monitoring changes in reservoir due to production. Using amplitude vs. offset (AVO) inversion, we are able to estimate changes in elastic and petro-physical parameters in addition to the spatial mapping of reservoir changes. Unfortunately, the complex properties of viscous oil-sands outlines deficiencies in most commonly applied AVO inversion techniques. These algorithmic deficiencies cause errors in our elastic and petro-physical estimations. The purpose of this proposed project is to model the effect of viscous fluids on seismic data in order to either account for, or remove its effects to obtain a more accurate AVO inversion result. TO BE CONT’D

Deterministic Decision Making Using All Geological Realizations

Predicting the quantity of mineral (gold, coper, etc) or hydrocarbon (oil, gas) extracted from a given deposit is difficult. One concern is the uncertainty in the resources underground. A few initial wells or drill holes give us access to an estimate of the quantity of resource, but there is great uncertainty because sampling can be hundreds of meters or kilometers apart. There can only be one plan for extraction which necessitates that the plan be optimal from the beginning of the project; computer simulations can help greatly with this and improving the efficiency of these models through this research will ensure optimal extraction plans. Moreover, most existing strategies ignore uncertainty and this work aims to optimize extraction plans while accounting for uncertainty in the subsurface. This increases profitability, recovery and reduces environmental footprints.

Large Consumer-Generated Data Optimization and Prediction

The proposed research aims to target large-scale consumer-generated data to analyze, visualize, and make predictions out of. The data will be collected from the consumers to make assessments on their lifestyles, and will come in forms such as heart-rate variance, that is, being temporal data. Researchers with visual analytics background will apply new visualization techniques on the data in order to grasp the insights and improve the model to interpret the data. The research problem is to relate measures of stress, recovery and mindful activities to the data obtained. The purpose of the proposed research is to complement the data with subjective measures such as happiness, life stress, and mindfulness. Some of the main objectives are to find out whether increased awareness of stress/recovery is associated with a change in stress/recovery; or whether increased daily mindful activity is associated with a change in stress/recovery. Expected results involve finding out whether the new product will contribute to a change in stress/recovery, or to increased mindful activity.

Pando.js: a platform for scientific volunteer computing on the Web

Scientists need more and more computing resources to do research. At the same time, a huge amount of available computing resources is currently completely ignored in all the devices and computers we use to surf the Web. All that computing power could be used to make progress on research about cancer, climate change, epidemiology, etc. that affects all of humanity. The overall project goal is to make these computing resources available to scientists working on public research. The project consists in creating Pando.js, a volunteer computing platform in the form of a website the general public will visit to contribute computing power. The computing
tasks will run in a webpage, created automatically from a scientist’s program source code, and accessed through social media links. The outcome will be a more convenient platform both for scientists and the general public to exchange volunteer computing resources than existing alternatives.

Supporting Role and Request Conceptual Modeling with Ontology-Based Reasoning

This project involves creating a comprehensive method for developing and maintaining a Meta-model for the representation of Role and Request Modeling (R2M) constructs based on the notion of ontologies. R2M provides clear definitions and semantics for the constructs of conceptual models as well as a set of rules for guiding the modeling process.

The intern will provide an ontology-based Meta-model for modeling R2M constructs, and rules. She will provide a method for recognizing any update that is not consistent with the defined semantics for R2M and a method for ensuring that all rules are conformed by any given model. ModiViz, which is the owner of the intellectual property relating to R2M, will use the outcome of this project for having the ability to ensure consistency in the Meta-model, paving the way to the introduction of much more robust model validation and guidance in the tool, and educating modelers much quicker.

Support à la conception de robots souples

La complexité de conception de robots souples est due à leur nature pluridisciplinaire (intégration de systèmes mécaniques, électronique et logiciel) ainsi qu’aux contraintes introduites par la non-rigidité des membres, ce qui met donc à rude épreuve la créativité des ingénieurs. Cette complexité est liée principalement aux multiples dépendances existantes entre les différentes composantes du robot déformable. Les multiples dépendances d’un système résulte généralement (dans les systèmes rigides) en l’adoption par les ingénieurs d’une conception séquentielle alors que la communauté de design conseille l’utilisation d‘une méthodologie intégrée (conception concurrente des sous-systèmes) car elle permet de réduire le coût de développement tout en améliorant la performance du robot. Différentes méthodologies ont récemment été développées pour faciliter cette conception intégrée des systèmes rigides, par contre, ces méthodologies ne considèrent aucunement les caractéristiques liées aux systèmes robotiques souples. Afin d’y remédier, une nouvelle méthodologie basée sur l’utilisation d’agents intelligents artificiels (AIA) sera développée pour fin d’aide à la décision au stade de la conception de systèmes robotiques souples. Cette méthodologie sera conçue sur des bases similaires à ce qui est utilisé dans la conception logicielle. TO BE CONT’D

Learning Approaches to Graph Structure Discovery

The project would explore the use of modern techniques from the field of Machine Learning to identify networks from observational data. This is an important area of research in fields such as neuroscience and genetics, where it can shed light on the nature of various disease. Other applications include discovering influencers and communities in social networks. These problems are normally untenable in the general case and techniques rely on deriving mathematical approximations, which often make many assumptions, and require lengthy time to formulate by expert researchers. In this project we explore methods of automatically deriving approximations to these problems based simulated examples, which are much easier to formulate than mathematical approximations. We hope the outcome of the project can provide
useful tools for researchers studying neuroimaging, genomics, and social networks to solve problems in their field.