Undergraduate Research Assistantship (URA) program

Computer networks have grown into a massive, shared infrastructure, connecting millions of users to thousands of online services. To support all these users and services, networks rely on a multitude of algorithms and protocols, e.g., for packet scheduling, congestion control, routing, load balancing. Together, these algorithms and protocols determine the quality of service that a network can provide to traffic flows for different applications in the network.

Understanding how different algorithms and protocols and their combinations affect network performance, finding the right set of algorithms and protocols given the specific requirements of the applications using the network, and creating robust and efficient frameworks for deploying them in the network are important open problems. We are looking for URAs to help us make progress on these problems by studying, analyzing, implementing, and experimenting with some of the most influential existing networking algorithms and protocols. Prior experience in systems-oriented courses such as Operating Systems and Computer Networks is a plus.

If you are interested, please fill out the Google Form at the following link: https://forms.gle/jvtgp7BdkaVFh3o39

The objective of this project is to reformulate the IESO’s algorithm to balance supply and demand based on societal gain. Societal gain as defined by this project is the cumulative difference between the carbon intensity (kg CO2e/MWh) consumers are willing to accept for the electricity consumed and that at which generators can produce that electricity. The solution to the reformulated optimization problem will present a different mix of generators and allow the calculation of an implied carbon tax required to achieve the least carbon intensive solution. Another objective of the project is to perform a sensitivity analysis by constraining the solution to varying maximum electricity prices.

The role for the URA student in this project is to take a leading role in the development of the code for the reformulated algorithm. The URA student will work directly with the involved graduate student and supervisor, and will be expected to perform independent research on the IESO market rules. The URA student’s efforts will contribute to the involved graduate student’s thesis.

The student would ideally have a background in CS, or in other programs with a focus on computational skills. The student is expected to master either Matlab or Python, and their optimization packages. The student is expected to be able to work independently and as part of a team. They must be highly motivated, goal-oriented, comfortable with high degree of freedom in their daily work and have excellent written and oral communication skills. The student is expected to have an interest in power grids and utility distribution, or to be willing to learn about them.

If interested, please email Robert Guglielmi.

We are witnessing an explosion of large language models (LLMs). The latest models such as GPT-4 have achieved unprecedented performance in various tasks such as code generation, text classification, and problem reasoning. However, serving LLM-based applications, i.e., deploying trained LLMs on a compute cluster and conducting model inference for incoming user requests, presents challenges in system design. GPT-4, for example, can take seconds to tens of seconds to generate responses, causing suboptimal user experiences for many interactive applications. Second, serving LLMs requires expensive hardware accelerators (e.g., GPUs) and results in high resource consumption. For instance, ChatGPT is estimated to cost over $700,000 per day to operate. Moreover, scaling LLM-based applications to meet the demands of large user bases can be challenging. Deploying the current ChatGPT model to power every Google search would require a cluster of over four million A100 GPUs. To address these problems, this project aims to design an LLM serving system with high utilization, low latency, and good scalability.

Interested students are expected to have programming capabilities (C, C++, and Python) and systems background (e.g., CS 350). Familiarity with LLM structures (or other ML models), and programming capabilities with TensorFlow or PyTorch can be a plus. For more information, please contact honzhang@uwaterloo.ca with your CV/resume.

I am interested in working with (multiple) undergraduate students on the topic related to security of machine learning. Students will implement a brand-new idea to conduct training/test-time defenses and attacks of machine learning, such as adversarial attack, backdoor attack, watermarking, poisoned attack, etc.

Working with team members and me, the aim is to publish a paper in a top ML or security conference. Students with combined machine learning and security/cryptograph/encryption backgrounds are highly encouraged to contact me and apply.

Excellent students will get my strong recommendation letter and the opportunity of joining my research team for MMath and Ph.D. study. There might be weekly discussions.

Prerequisites: Experience with Python and PyTorch. Received a high grade in CS480/680. Basic understanding about AI security.

For more information, please contact Professor Hongyang Zhang at: hongyang.zhang@uwaterloo.ca.

This project focuses on data-driven approaches to secure 5G network services (i.e., eMBB, mMTC and URLLC). In this project, the URAs will be tasked to generate realistic datasets for benign 5G traffic and varying attack scenarios that could lead to compromised 5G services. The datasets generation will be carried out in our emulated and/or in-lab 5G testbed environment. The URAs will also leverage generative adversarial networks to up-scale the datasets, while ensuring that the statistical distribution of the generated datasets is preserved. Furthermore, the URAs will explore machine/deep learning techniques for early detection of attacks, including zero-day attacks, on 5G services using the generated datasets.

Interested students should contact Raouf Boutaba at: rboutaba@uwaterloo.ca with an unofficial transcript and a resume.

In mobile networks, radio access networks (RANs) have been traditionally deployed in vendor-specific and closed architectures. This makes it difficult to manage and control the network and integrate new services. To address these, Open RAN, a service-oriented software-based architecture has been proposed based on the softwarization of different RAN elements. Open RAN introduces the radio intelligence controller (RIC), a centralized abstraction that allows for enhancing traditional RAN functions with intelligent closed-loop control powered by machine learning (ML). To this end, customized RAN control applications (called xApps) are deployed on RIC and communicate with RAN components to manage different aspects of the network including the radio resource allocation. The student is expected to study existing open-source RIC frameworks and contribute to the development of xApps and ML algorithms for RAN control. The student will also have the chance to integrate and test xApps with our in-lab 5G testbed.

Interested students should contact Raouf Boutaba at: rboutaba@uwaterloo.ca with an unofficial transcript and a resume.

Simulations of complex wave phenomena brings together fascinating topics in physics and computer science. Such simulations have many diverse applications: like simulation of electromagnetic radiation, for example for propagation of RADAR or cellular signals in real-world environments; simulations of optical medical devices; as well as accurate light transport and rendering in computer graphics.

This project involves research and development of relevant tools, implementing simulation software (like path tracers), and performance analysis of algorithms or methods. Students are expected to have strong programming skills, and ideally some knowledge of computer graphics (e.g., CS488), optics or electrodynamics. Interested students should contact Shlomi Steinberg (steinberg@uwaterloo.ca).

In this project, a student will assist our research group to collect fundamental data and scientific information for our ongoing research on variations of light transmission through snow elicited by climate changes, and their impact on the vegetation cover of boreal and alpine regions. Moreover, the student will also conduct computer simulations related to this research using the online version of our spectral light transport model for snow.

Interested students should have experience with data processing and plotting in Matlab. For more information, please contact Prof. Gladimir Baranoski at:gvgbaran@gmail.com

It is a well-known fact that a well-designed light transport model is of little use without reliable supporting datasets. These include biophysical data (e.g., refractive indices and absorption coefficients) and specimen characterization data (e.g., thickness and pigment concentrations) to be used as input, as well as evaluation data (e.g., reflectance and transmittance curves) to be used in the assessment of its predictive capabilities. Although one may assume that such datasets are readily available, especially for ubiquitous materials such as plants, human tissues and soils, in reality this is not case. In fact, a common constraint shared by all these datasets is scarcity. Oftentimes, the search for such datasets becomes the bottleneck of research projects. The Natural Phenomena Simulation Group (NPSG) is looking for undergraduate students interested in acquiring hands on experience on this important aspect of scientific research and in contributing to some of our projects, which include, for example, the detection and analysis of subsurface targets in terrestrial and extraterrestrial terrains and the screening of medical conditions.

Interested students should be familiar with fundamental numerical analysis concepts and have experience in using Matlab. Experience with applied optics and computer graphics will be taken into account, but it is not required.

Interested students should contact Prof. Gladimir Baranoski at: gvgbaran@gmail.com

Cryptography can protect data in rest and during transit. However, keeping data encrypted during computation is still often too inefficient. There exist cryptographic techniques such as homomorphic encryption and secure multi-party computation but they require careful application in order to be practically efficient. We work on several projects that apply these techniques, such that the computational overhead is reduced, e.g., by performing some computations on plaintext data, combining different cryptographic primitives or optimizing the protocols to reduce the performance bottleneck, e.g., communication cost. We focus on applications for big data, such as private set intersection, private information retrieval or private inference over data models, because these applications are already deployed in industry. We build cryptographic protocols using different toolkits and libraries, such as Microsoft's SEAL or MP-SPDZ, which are usually written in C++.

If you are interested in this URA, please email Florian Kerschbaum (florian.kerschbaum@uwaterloo.ca) with an unofficial transcript and a resume.

More and more applications use machine learning to derive insights from large data collections. However, this process is susceptible to several security and privacy threats. For example, the data collection may contain sensitive, private information that may still be derived from the model or the learning and inference process. We work on several projects that help ensure that such threats are contained. We work on devising improved attacks that demonstrate that protection mechanisms are not as successful as they claim to be or processes that are assumed to be safe are not. We also work on defense mechanisms that provide better protection based on the latest developments in cryptography, differential privacy, and machine learning. Our work involves designing algorithms, developing prototypes, mostly in Python, and evaluating their performance and security.

If you are interested in this URA, please email Florian Kerschbaum (florian.kerschbaum@uwaterloo.ca) with an unofficial transcript and a resume.

ICLab is a global and longitudinal Internet censorship measurement platform that collects data on censored webpages around the world since 2016. ICLab makes use of commercial VPNs as vantage points on multiple countries to detect censorship activity based on DNS manipulation, TCP packet injection, and the serving of “blockpages”.

Currently, however, ICLab is only able to archive raw measurement data for posterior analysis and lacks the ability to provide non-technical users with a data visualization dashboard that facilitates an easy exploration of the platform’s measurements as new data is collected.

The goal of this project is to design and develop an ICLab visualization dashboard, based on web technologies, that enables a user-friendly exploration ICLab measurement data (e.g., at a country and measurement type granularity).

Interested students should have practical experience with information visualization technologies (e.g. React-D3, Plotly Dash, etc.) and relational databases (SQL). For more information, please contact Prof. Diogo Barradas at dbarrada@uwaterloo.ca and send an unofficial transcript and a resume.

If you are a senior undergraduate student interested in working with large scale distributed systems that involve managing big data and storage in the cloud to achieve scalability and deliver good performance, please e-mail Khuzaima Daudjee at: kdaudjee@uwaterloo.ca

Data privacy is becoming one of the major challenges of the tech industry. With many applications outsourcing their data storage needs to third party cloud vendors such as Amazon AWS or Google Cloud Platform, the lack of trust in third party storage services aggravates the privacy challenges. While encrypting an end user's data forms the first layer of protection against the loss of data privacy, many recent attacks have shown that the access patterns on encrypted data reveal sensitive information. For example, the duration and frequency with which an oncologist accesses (encrypted) data can reveal the type of a patient’s cancer (e.g., based on the frequency and intervals of chemotherapy treatments). In this project, we will explore efficient techniques to hide the data access patterns from an adversary and integrate these techniques in building a privacy-preserving datastore.

A URA candidate would begin the project by reading the necessary background research papers, will be actively involved in research meetings, and will be involved in implementing the newly designed data system

If interested, please email Sujaya Maiyya at: smaiyya@uwaterloo.ca

Familia is a language design that promises a lot of polymorphism and extensibility in a lightweight, type-safe package. This project is aimed at materializing all this expressive power by creating a compiler for Familia. In particular, students will implement compiler passes transforming Familia code into lower-level languages.

We are seeking ambitious students who have compiler-hacking experiences and, ideally, are familiar with LLVM.

Interested students should contact Yizhou Zhang: yizhou.zhang@uwaterloo.ca

Recent advancements in ML techniques, such as large-language models (LLMs), revolutionized natural language processing and understanding. At the same time, these large-scale ML applications pose new challenges. One of the main aspects is the demand for computation capabilities. For example, serving LLM costs more than ten times as compared to conventional web search. This URA project aims to improve the performance of large-scale ML applications through hardware support, such as designing accelerators for ML applications and optimizing their memory systems.

Interested students are expected to have programming capabilities (C, C++, and Python) and computer hardware and systems background (e.g., CS 350 and 450). For more information, please contact sihangliu@uwaterloo.ca, and send an unofficial transcript and your CV/resume.

If you are a senior undergraduate student interested in working with large scale distributed systems that involve managing big data and storage in the cloud to achieve scalability and deliver good performance, please e-mail Khuzaima Daudjee at: kdaudjee@uwaterloo.ca