Four students at the Cheriton School of Computer Science are recipients of the Computing Research Association’s 2024 Outstanding Undergraduate Researcher Awards. The annual CRA awards program recognizes students from universities across North America who have distinguished themselves by conducting exceptional computer science research as undergrads.
This year, Matthew Yang was a finalist, and Ruidi Wei, Jiawen Zhu and Alex Zhuang each received honourable mentions for their research.
“Congratulations to Matthew, Ruidi, Jiawen and Alex on receiving these recognitions from the Computing Research Association,” said Raouf Boutaba, Professor and Director of the Cheriton School of Computer Science. “Through their publications, they have truly distinguished themselves as exceptional young researchers amongst a very competitive pool of research-oriented undergrads in universities across Canada and the United States.”
The Computing Research Association is supported by Microsoft Research and Mitsubishi Electric Research Labs, which sponsor the Outstanding Undergraduate Researcher Award program in alternate years. This year, Mitsubishi Electric Research Labs sponsored CRA’s Outstanding Undergraduate Researcher Awards.
Matthew Yang | Advised by Professors Gautam Kamath, Florian Kerschbaum, Jimmy Lin and Yaoliang Yu
“Last summer I had an idea for a project,” recounts Gautam Kamath, one of three professors who supervised Matthew’s research. “With my colleague Yaoliang Yu and his student Yiwei Lu, we had some promising results on understanding indiscriminate data poisoning attacks. In these settings, an attacker adds datapoints to the training dataset with the goal to decrease the test accuracy as much as possible. We wanted to scale up our investigations to larger and more common settings in practice. I proposed including Matthew in our next steps and he gladly agreed.”
“We were travelling at the time, so I asked Matthew to catch up by reading previous papers and playing around with the code,” Professor Kamath continued. “Despite this loose guidance, he re-implemented our previous methods from scratch, which were even more performant than before. From that starting point, we worked together to produce a paper that has been conditionally accepted to SaTML 2024, the IEEE Conference on Secure and Trustworthy Machine Learning.”
Matthew also worked as an undergraduate research assistant with Professor Florian Kerschbaum, whose research is in data security and privacy.
“Matthew and I worked on differentially private nearest neighbour queries,” Professor Kerschbaum said. “The nearest neighbour problem is common in data science and asks to find the closest data points in a database to a query data point. We investigated this problem under a privacy constraint where the query data point is not to be revealed to the database management system, using a modern notion of privacy known as differential privacy. The challenge of Matthew’s research was to evaluate this trade-off and compare our protocol to state-of-the-art protocols. Although we worked together for just one term, I was immensely impressed by Matthew’s research capabilities. He has excellent communication skills and is a truly curious, thorough and methodical undergrad researcher.”
Matthew also conducted research with Cheriton Chair Jimmy Lin, whose research is in information retrieval and natural language processing. During his time with Professor Lin, Matthew published two papers, one of which was as first author and presented at the 2022 ACM/IEEE Joint Conference on Digital Libraries and a second as a coauthor published in ACL 2023, the Findings of the Association for Computational Linguistics.
“Matthew has distinguished himself through his outstanding accomplishments in research, making significant contributions to works across a broad range of areas during his undergraduate studies at Waterloo,” Professor Kamath said. “He has impressed everyone along his path.”
Ruidi Wei | Advised by Professor Florian Kerschbaum
“Ruidi approached me with the desire to conduct research, so I proposed the rough sketches of the idea that should become our publication,” said Professor Kerschbaum. “In his first part-time term, I provided Ruidi with the necessary background reading to understand the problem space. I was impressed by Ruidi’s capabilities to absorb knowledge quickly and deeply. He always had insightful questions during our discussions of this material and quickly understood advanced concepts.”
In their research, Ruidi and Professor Kerschbaum considered the record linkage problem under privacy constraints. Private record linkage is a problem where two parties wish to determine the intersection of their data, but their records may slightly differ because of data errors or schema mismatches, and they do not want to disclose anything outside the intersection. They set out to answer the open question whether a cryptographically secure protocol — one without additional leakage but using a locality-sensitive hash function — promises higher accuracy than its alternatives.
“Ruidi first had to construct an efficient method for an oblivious join using secure multiparty computation, which is at the core of this contribution,” Professor Kerschbaum said. “He modified the best-known method from my former master’s student Simeon Krastnikov, which was designed for trusted execution environments. Ruidi came up with an implementation, but I challenged him to provide a proof of correctness, which he produced within a week. He then implemented the entire algorithm in a multiparty computation framework, namely ABY which was developed by the Technical University of Darmstadt. Ruidi had to find many ways to optimize the program, including finding the optimal secret share encoding for each computation, considering the costs of conversions between encodings, and performing local computations, which are not supported out of the box by ABY.”
In his second full-time research placement, Ruidi investigated the question of accuracy. Here he conducted another full implementation but without the cryptographic guards since they slow down the evaluation too much.
“Ruidi prepared data sets that we chose from the Internet, performed experiments, and generated the resulting graphs,” said Professor Kerschbaum. “The result of Ruidi’s implementation, optimization and evaluation is a protocol that outperforms the state-of-the-art not only in security, but also in efficiency and accuracy. It is a truly great achievement and acceptance of this impressive result by VLDB is the direct consequence of Ruidi’s insights and hard work.”
Jiawen Zhu’s research | Advised by Professor Jian Zhao
“I knew Jiawen from a course I taught in spring 2022 — CS 449/649 Human-Computer Interaction,” recounts Professor Zhao. “With her strong motivation and excellent performance in my class, Jiawen joined my lab as an undergraduate research assistant, where I continue to advise her.”
She has been working under Professor Zhao’s supervision for more than a year now, and has contributed to a number of research projects.
For example, one project Jiawen participated in examined affective communication when voice messaging with smartwatches, working with a postdoctoral researcher and several undergrads. In face-to-face communication, people can use non-verbal information such as facial expressions to predict the emotional tone of the other party. However, because of the nature of voice messaging, users cannot gauge the content of the messages before listening to them.
“In this project, we explored the concept of emotional teasers, by adding animations or colours to message bubbles,” said Professor Zhao. “We compared the two types of teasers through a user study. Jiawen helped run pilot studies and iteratively improved the study task and questionnaire design. She also conducted the user studies and analyzed and visualized the collected quantitative data to understand user perception of emotional teasers. She drafted the quantitative results section of the manuscript, a full paper that has been submitted to CHI 2024, the top conference in human-computer interaction.”
Moreover, Jiawen worked on supporting the prompt authoring process when generating code using AI-powered code assistants. With the development of large language models, code assistants are able to generate large amounts of high-quality code. However, current tools lack controllability so programmers must spend considerable time understanding and editing the produced code if their intended outcome is not achieved. To address this problem, a system was proposed that provides hierarchical prompt authoring to give users more fine-grained control of the code-generation process.
“Jiawen ran formative interview studies and performed thematic analysis on the resulting qualitative data to distill principles that can inform our design,” Professor Zhao said. “She prototyped the interface and implemented parts of the system. She also ran evaluation studies to determine the usefulness of the proposed system. Jiawen analyzed the quantitative data, and drafted the study procedure and quantitative results sections of the manuscript, which has been submitted as a full paper to CHI 2024.”
As Jiawen has gained considerable knowledge and skills in HCI research since fall 2023, she has started to conduct a new research project independently.
“Her project aims to invent new technologies that address the unique challenges faced by grandparents and grandchildren in immigrant families when struggling to connect with each other because of language barriers, cultural differences, and geographic distance,” Professor Zhao said. “I believe that Jiawen can lead this project and generate fruitful results.”
Alex Zhuang | Advised by Professor Wenhu Chen
“Alex’s talent for research is truly remarkable. In our collaboration, he demonstrated not only exceptional technical prowess but also a keen intuition for identifying and addressing meaningful problems in natural language processing,” Professor Chen said. “Alex and I co-authored a paper with Tianle Li, Xueguang Ma, Yu Gu, Yu Su titled Few-shot in-context learning on knowledge base question answering.”
This research examined question answering over knowledge bases, a difficult problem because of the challenge of generalizing to a wide variety of possible natural language questions. The heterogeneity of items across different knowledge bases often necessitates specialized training for different knowledge base question-answering datasets. To handle questions over diverse knowledge base question-answering datasets with a unified training-free framework, the research team proposed KB-BINDER, which for the first time enables few-shot in-context learning over knowledge base question-answering datasets tasks. The experimental results on four public heterogeneous knowledge base question-answering datasets revealed that KB-BINDER has strong performance with only a few in-context demonstrations.
“This
paper
has
been
accepted
at
the
annual
Association
for
Computational
Linguistics
conference,
which
speaks
volumes
about
the
quality
and
significance
of
Alex’s
contributions,”
Professor
Chen
said.
“Beyond
his
technical
accomplishments
to
this
work
what
stands
out
about
Alex
is
his
unwavering
dedication
and
work
ethic.
He
is
diligent
and
consistently
proactive
in
his
approach
to
research.
These
qualities,
coupled
with
his
passion
for
natural
language
processing,
make
him
an
invaluable
asset
to
any
research
project
or
team.”