Undergraduate Research Fellowship (URF)

Supervisor: Florian Kerschbaum

The development of private computation techniques increases opportunities for two or more companies to use their data collectively for analysis without disclosing private user data to one another. However, the use of private computation does not absolve companies from their responsibility to disclose their data practices to users. We aim to ensure users, who are the data subject and have provided their data to a company, can give informed consent. We intend to study the expectations users have as data subjects for companies that include their data in private computation. Our results could guide law and policy surrounding how companies communicate to users about private computation as well as provide researchers with a better understanding of users’ expectations for privacy in this setting.

Supervisor: Hongyang Zhang

The project aims to simulate daily flows by deploying the first algorithm to an individual flight’s user preferred route. This is a planning exercise conducted on behalf of each airline/each flight, usually 3 hours prior to takeoff. To this end, planned upper meteorological data (licensed by Transport Canada) will be used. This data covers the entire globe and over 15 altitudes used by commercial aircraft, uses 36-hour ‘look ahead’ forecasts, and is updated 4 times daily. The objective is to design a reliable machine learning algorithm that analyzes the entire planned traffic sample flying each day across the North Atlantic (NAT). Tasks include: 1) Revising and re-assembling current 4D shortest path algorithm; 2) Parallelizing optimization runs to accommodate simultaneous flight computations; 3) Investigate the robustness, security, and reliability of the algorithm; 4) Investigate machine learning using model Planned versus Flown data sets. An ideal URF should have basic knowledge on machine learning and algorithms (e.g., the shortest path algorithm) and strong coding ability. Besides money support, successful URFs will get my recommendation letters or the opportunities to join my lab as graduate students.

Supevisor: Tim Brecht

The NHL collects puck and player tracking data during hockey games. This data provides significant opportunities for new advanced hockey analytics and user engagement. The goal of this project is to develop analytics that can be used by players, coaches, general managers, broadcast analysts, fans and gaming/betting sites as well as in the development of advanced models of players and teams. (preliminary research)

Supervisor: Martin Karsten

Operating system kernels are fairly big and very complicated software entities that address complex resource management challenges and typically support a massive set of hardware devices. Meaningful research into the structure and performance of operating systems is hampered by a significant barrier to entry: A research operating system must support a reasonable set of modern hardware devices to obtain useful performance measurements beyond simplistic benchmark tests. The overall goal of this project is lowering that barrier to entry by building a simple kernel nucleus and combining it with 3rd-party open-source software to support a large variety of device drivers. The critical next step is hollowing out an existing open-source operating system kernel and making the hardware support components independent of the core generic resource management services. This will result in a novel open-source research platform that enables subsequent studies on structural and algorithmic innovations for operating system kernels.

Supervisor: Toshiya Hachisuka

Statistical approaches have been used in computer graphics for many different problems. In particular, in light transport simulation for photorealistic image synthesis, stochastic estimation using Monte Carlo methods is the most popular and efficient approach. This project aims to investigate mathematically and numerically the effectiveness of specific statistical methods for problems in computer graphics and to discover more efficient computational methods. The student is expected to study existing statistical methods such as Monte Carlo methods, experiment with them by coding a test program in computer graphics and work on its mathematical formulation to understand why and how well it works on a problem in computer graphics.

Supervisor: Jian Zhao

Children with autism spectrum disorder (ASD) face many difficulties in daily circumstances, due to their impaired social and communicative functions as well as restricted interests in people and activities. Emerging studies have shown that neurofeedback training (NFT) games are an effective and playful intervention to enhance social and attentional capabilities in children with ASD, leading to improved behavior indicators. While existing studies have examined the effects of NFT games as interventions for ASD, current NFT games still have many limitations. This project aims to investigate novel interaction techniques (e.g., augmented reality, mobile sensing) for enhancing traditional NFT games with a broader range of other interventions. The outcomes of the project will lead to more engaging, accessible, and personalized NFT-based games for the large population of autistic children. An ideal candidate should be proficient or knowledgeable in game and graphics design, mobile platform programming (e.g., iOS), and basic machine learning.

Supervisor: Christopher Batty

Accurate computer simulations of the physics of liquids are important for a wide range of disciplines and applications: lava flows and splashing waves in visual effects-driven films; biological processes in the body such as digestion and blood flow; atomization processes in which a high speed jet of liquid disintegrates into a dispersed spray; and many more. A very effective way to represent the surface geometry of a liquid is as a deforming, wireframe-like mesh of connected triangles. However, while this method produces extremely accurate results, it remains quite slow. In this project, we will investigate strategies to significantly accelerate such methods, such as improved collision detection algorithms, more memory-efficient data structures, and multicore or GPU parallelism.