Three Honorees Find Inspiration, Validation from David J. Kuck Outstanding Thesis Awards

11/11/2022 2:57:41 PM Aaron Seidlitz, Illinois CS

Since 1996, Illinois Computer Science has provided the David J. Kuck Outstanding Thesis Awards to a PhD and master’s student in recognition of the former professor’s intellectual and leadership contributions. This year, honorees included:

  • Apostolos Kokolis, who graduated from the PhD program in 2022
  • Qi Zeng, formers master’s student who is now a PhD candidate
  • Yuheng Zhang, former master’s student who is now a PhD student

The awards were established by alumni, former students, and friends of Kuck, who was a professor of computer science from 1965-1993. He was founder of Kuck and Associates and won numerous awards, including the Eckert-Mauchly Award from ACM/IEEE, the IEEE Computer Society's Computer Pioneer Award, the Charles Babbage Outstanding Scientist Award, and the Illinois Computer Science Distinguished Educator Award. In 2015, Kuck was inducted into the Grainger College of Engineering Hall of Fame.

Congratulations to each of the winners for this year’s David J. Kuck Outstanding Thesis Award:

Apostolos Kokolis, David J. Kuck Outstanding PhD Thesis Award
Thesis: New Architectures for Non-Volatile Memory Technologies

Apostolos Kokolis
Apostolos Kokolis

Kokolis came to Illinois CS after studying Electrical and Computer Engineering in Greece, where he was inspired by the ability to interact with other fields through computing.

Now, looking back, Kokolis said there were three primary reasons he enjoyed studying here. First, he was impressed by his PhD advisor Josep Torrellas’ dedication to computer architecture and problem solving. Kokolis also enjoyed the freedom Torrellas provided him to address research problems specific to his own interests in computing. Finally, the entire experience provided by Torrellas during the PhD program provided Kokolis with a necessary balance between his own inspiration for research and a guiding presence that helped him define problems accordingly and challenged him to find the right solutions.

That process led to his PhD thesis, which Kokolis worked on after realizing that “there was an ongoing shift in Computing Memory Systems with the introduction of Non-Volatile Memory (NVM).”

“This new memory type showed promise to satisfy the memory needs of emerging applications and the needs of vast, fast storage,” Kokolis said. “In my thesis, we started by introducing new single-node architectures that incorporate NVM to achieve high performance and enable the programmability of NVM systems. Besides single-node systems, distributed/cloud systems are of great importance. Companies such as Google, Microsoft, and Meta need to satisfy the increased demand for data storage and low data access latency in their cloud services.

“In this area, we found that, although there are years of research around consistency models, there is no framework for establishing and evaluating the different distributed persistency models. This led to the ‘Distributed Data Persistency’ work that opens this new area of research for distributed systems and continued with our work on distributed transactions.”

This latter portion of the work is what he considers the biggest contribution of his thesis, as Distributed Data Persistency defines numerous persistency models and provides a comprehensive framework that “captures the performance, durability, and programmability/intuitiveness of such models.”

Kokolis believes there is a way for this work to impact both research and industry moving forward.

The entire process, he said, was challenging but rewarding – which is why he was honored to receive the David J. Kuck Outstanding Thesis Award.

“It is very rewarding to be the recipient of this award. The PhD is a long process that naturally has many difficulties to overcome along the way. It is very fulfilling to realize that the work of so many years is recognized and appreciated,” Kokolis said.

Qi Zeng, David J. Kuck Outstanding MS Thesis Award
Thesis: Event Network Embedding

Qi Zeng
Qi Zeng

Master’s student Qi Zeng said that her inspiration for researching Natural Language Processing, especially Document-level Language Understanding, stemmed from an internship spent working on a chatbot called Microsoft XiaoIce.

Her interest grew because it refreshed “my impression that words, emotions, and intents can be computed.” As Zeng realized her interest, she began looking at who she wanted to study with at the graduate level. At that time, she had already heard of Illinois CS professor Heng Ji.

So, when Ji offered her the opportunity to study here together, she was thrilled to accept the invitation.

Zeng narrowed her interest to Document-level NLP, which eventually became the work that formed her awarded master’s thesis. By working on global event network embedding, Zeng tried to represent events in a more global context other than sentences.

“Then I found the broad applications and remaining challenges for the document-level settings, such as efficiency and consistency problems. This confirmed to me that this was a topic worth researching,” Zeng said. “At that time, we were looking at a more efficient way to represent an event. Previous methods treated events as tuples and ignored their context.

“I tried to use the view of graphs to represent events and borrowed ideas from network embedding. We have some novel model designs since events, as the processed unit, have their own unique characteristics. Also, because event representation is a relatively new area, we designed the evaluation benchmark by ourselves.”

Now with the thesis finished, Zeng said the most important impact of the work should be proving that network embedding techniques can be used in event representations.

It’s a meaningful result that stemmed from her time working alongside Ji.

“Professor Ji’s high-level advice on problem formulation and experiment design is most helpful to me. This was my first independent project, so I really learned a lot from her,” Zeng said. “And I am super grateful for the Outstanding Thesis Award, because this is the first award I received after joining the graduate program at Illinois CS. It means a lot to me to be recognized this way for the first time.”

Yuheng Zhang, David J. Kuck Outstanding MS Thesis Award
Thesis: Active Heterogeneous Graph Neural Networks with Per-Step Meta-Q-Learning

Yuheng Zhang
Yuheng Zhang

Yuheng Zhang had a vision to join Illinois Computer Science as a master’s student with the intention of leveraging this experience into a PhD program. His research focus on active learning and its connections to graph mining and reinforcement learning, all conducted through work with Illinois CS professor and his advisor Hanghang Tong, positioned Zhang to do exactly that.

“This all began from my first meeting with professor Tong, in which we talked about his vision for the human-agent team – or the interaction between humans and intelligent agents,” Zhang said. “Active learning is an important part of it, and the idea is very exciting to me. That is why I decided upon active learning as my research focus.”

His inspiration stemmed from the idea he and Tong had to “transfer knowledge of active learning from source tasks to target tasks.”

They did so by designing a new framework based on techniques from a paper on meta-q-learning.

The biggest impact from the work, Zhang believes, is that the thesis does indeed provide a new way to think about active learning problems.

“We can view it as a meta-reinforcement learning problem and exploit the knowledge from source tasks to train a more effective active learning strategy,” Zhang said. “Working with professor Tong always provided insightful thought on our research area. When I was stuck by a problem, he could always point out the right direction to proceed. I learned a lot from our meetings. Besides research projects, we also have set aside teatime for casual chatting. I got a lot of encouragement and help from him.

“And winning the Outstanding Thesis Award brings me great motivation and encouragement. It makes me more confident to study more challenging research problems. I hope that I can solve some fundamental machine learning problems in the future.”