As Illinois CS research continues to sustain high levels of success, one indicator is that the papers produced here earn best paper awards at some of the most prominent computing conferences.
From the faculty guiding these projects to the students learning what it takes to develop well-regarded research, these honors are representative of the effort and dedication given to research at Illinois CS.
Congratulations to each of the following projects and the authors involved.
Top Pick, IEEE Micro Special Issue
Illinois CS professor Josep Torrellas, and his i-acoma research group, earned two Top Pick selections for the IEEE Micro Special Issue. This selective distinction honors 12 papers out of all the published pieces in the computer architecture research area over the past year.
The two papers honored as a Top Pick include:
- Distributed Data Persistency, co-authored by Torrellas, Illinois ECE professor Jian Huang, as well as Illinois CS students Apostolos Kokolis, Antonis Psistakis, and Benjamin Reidys.
- Maya: Using Formal Control to Obfuscate Power Side Channels, co-authored by Torrellas, Alexander Schwing – Illinois CS affiliate faculty member and professor with Electrical & Computer Engineering – and University of Nevada professor Petros Voulgaris, as well as Raghavendra Pradyumna Pothukuchi, Associate Research Scientist in the Deptartment of Computer Science at Yale University, and Illinois CS student Sweta Yamini Pothukuchi.
The inspiration for “Distributed Data Persistency,” according to Torrellas, comes from three factors that are contributing to major changes coming to data centers.
“First, networking hardware is becoming faster, reducing the inter-node communication latency. Second, new Smart Network Interface Cards (SmartNICs) are progressively off-loading more computation from the CPUs. Third, non-volatile memory (NVM) is about to become widely used in datacenters; this memory substantially reduces the cost of making data durable compared to solid-state disks (SSDs),” Torrellas said.
This group calls the combination of two models – the data consistency model and the memory persistency model – the Distributed Data Persistency (DDP), which they believe will be crucial to ensuring correctness and performance of distributed applications – like those developed by Google, Meta, and Amazon.
“Providing an understanding of the interaction between data consistency models and memory persistency models helps programmers of cloud applications and hardware designers,” Torrellas said. “Programmers can now understand the tradeoffs between performance, reliability, programmability, and ease of implementation of different models.
“Hardware designers, additionally, know how to build all these models. We have heard of at least two companies studying our work.”
In regards to “Maya,” Torrellas said this group spent three years on a “revolutionary new idea” that helps offset one of the “most vexing challenges, as we try to protect our cyber infrastructure.”
This problem came down to leaking information through physical channels (power consumption, temperature, and electromagnetic emanations), which attackers measure and use machine learning techniques to predict which programs, passwords, or algorithms are used.
“We thought that rigorous techniques based on control theory might be a good way to obfuscate the power consumed by a program,” Torrellas said. “Our techniques have already been used by other researchers to defend against attacks. Specifically, the attack consisted of having multiple computers connected to the electrical outlets in a given building, and passing information between them through power spikes. Our work neutralized this attack.
“Another impact of our techniques is that more researchers are developing defenses against power attacks.”
Top Pick, IEEE Micro Special Issue
“A labor of love” over the last several years has led Illinois CS professor Sarita Adve, and a student research team – including graduate student lead Muhammad Huzaifa – devoted to the Illinois Extended Reality (ILLIXR) testbed and consortium, to a Best Paper Award at the IEEE International Symposium on Workload Characterization (IISWC 2021) and now a Top Pick in the IEEE Micro Special Issue.
This paper, entitled “ILLIXR: Enabling End-to-End Extended Reality Research,” provided the initial description of the entity’s purpose. It does so, as Adve explained, by providing “the first published power/performance/quality characterization of an end-to-end extended reality system.” Additionally, it “provides a blueprint for the research my group will undertake over the next several years.”
Initially, Adve and this research team grew a bit discouraged after a couple rejections from top conferences. Simultaneously, the group was gaining momentum through more adoption with industry partners. While frustrating, as academic success often correlates to prestigious publications, the group remained steadfast in its belief.
Continued adoption of ILLIXR provided more depth to discuss its success, eventually leading to the IISWC Best Paper recognition.
The Top Pick designation forms the next step in much-deserved recognition of this effort.
“The top picks recognition means a lot. It is a very selective recognition delivered to the top twelve papers from all of computer architecture conferences in the last year – based on novelty and potential for long term impact,” Adve said. “Our paper is unusual in that it describes a new research vision for the architecture and systems community backed by a unique research testbed (the first open source end-to-end XR system), the first published characterizations of power/performance/quality of experience for an end-to-end XR system, and many implications for research for architecture, compiler, systems, and algorithms researchers.”
Best Demo Paper, 2021 North American Chapter of the Association for Computational Linguistics
A collaborative effort, orchestrated through Illinois CS professor Heng Ji’s team of experts who have worked with her on previous work funded through DARPA, earned the Best Demo Paper at the this year’s North American Chapter of the Association for Computational Linguistics (NAACL 2021)
The paper, titled “COVID-19 Literature Knowledge Graph Construction and Drug Repurposing Report Generation,” helps in the effort to combat COVID-19. As the paper’s abstract notes, “clinicians and scientists need to digest vast amounts of relevant biomedical knowledge in scientific literature to understand the disease mechanism and related biological functions.”
To help in this effort, the paper introduces a “novel and comprehensive knowledge discovery framework, COVID-KG, to extract fine-grained multimedia knowledge elements from scientific literature.”
Ji noted that others have already cited the paper 51 times since June last year.
“Our research provides a way to turn the massive unstructured literature into a structured knowledge graph. COVID-KG starts by reading existing papers to build multimedia knowledge graphs (KGs), in which nodes are entities/concepts and edges represent relations and events involving these entities, as extracted from both text and images. As a result, it can accelerate scientific discovery and build a bridge between the research scientists making use of our framework and clinicians who will ultimately conduct the tests,” said Manling Li, paper co-author and Illinois CS PhD student.
Li also credited another Illinois CS professor and co-author Jiawei Han, whose experience with biomedical projects and knowledge discovery was instrumental to the effort. Also, David Liem of UCLA, many Illinois CS students, and computer vision experts like Shih-Fu Chang all played major roles in this development. Other research partners came from Brandeis University, the University of Washington, Colorado University, DARPA and ARL.
“It was a great pleasure to see many people using COVID KG, as it supports or initiates a lot of downstream tasks – such as drug discovery, mortality prediction, etc.,” Li said. “Such system framework also provides detailed contextual sentences, subfigures, and knowledge subgraphs as evidence. It has been downloaded more than 2,000 times. The greatest pleasure comes from seeing how it benefits clinicians, scientists, and researchers in a variety of fields.”
Distinguished Paper Award, ACM SIGPLAN Symposium on Principles of Programming Languages (POPL 2022)
Illinois CS Professor Madhusudan Parthasarathy and student Paul Krogmeier worked together for about the past year and a half on a paper, “Learning Formulas in Finite Variable Logics,” that earned a Distinguished Paper Award at the 2022 Principles of Programming Languages (POPL) conference 2022.
The origin of the work went back to a previous paper of Parthasarathy’s, which he believed could be improved upon. And he felt that working with Krogmeier might just be the key to unlocking the potential, due to his pupil’s interest in machine learning and logic. This coincided with the professor’s recent interest in exploring synergies between machine learning and program synthesis for building intelligent systems.
“Humans learn from examples as well as learn to reason effectively in a variety of domains. The intersection of logic and learning is hence a very interesting area that our paper makes fundamental contributes to, but it would be good to see a lot more work in this area,” Parthasarathy said.
As stated in the paper’s introduction, the two authors “embark on a foundational study of exact learning of logical formulas.” Their belief is that symbolic expressions, “can be easily analyzed and interpreted, which aids downstream applications and makes them easier to communicate to both humans and computers.”
Parthasarathy shared what he believes to be the most beneficial finding from their work.
“A clearer understanding of when interpretable learning is possible, and the underlying decision procedure gives the algorithmic structure to build more practical interpretable learning and program synthesis,” he said.
To not only successfully work through the paper, but for it to earn this distinction at POPL ’22 became a significant achievement for Krogmeier.
“It’s encouraging to know the ideas and the presentation of the work are appreciated. I think the paper does a good job presenting a few simple and important ideas rather cleanly. We took care to make it that way, and I hope that any reader finds it pleasant,” Krogmeier said.
Excellent Paper Award, IEEE journal on Big Data Mining and Analytics
For the past 10 years, Illinois CS professor Hanghang Tong has kept a long-term research connection with a collaborative group from Nanjing University in China. The positive nature of this collaboration across many years allows for insight into several topics, as interest in something new comes up amongst members of the group.
With a strong working background intact, this group provided an award-winning paper entitled “A Brief Review of Network Embedding.” Tong said interest in “network embedding has become one of the most popular topics in graph learning and mining, with tens of – if not hundreds of – new papers each year.”
The resulting paper, Tong said, provides “a clear and concise taxonomy of network embedding, elucidating the key ideas of a few representative works in each category of network embedding methods. It helps the beginner to navigate in this vast and fast-growing field of network embedding.”
Tong and his group found a number of nice surveys already on the topic, but they believed most to focus on completeness or comprehensiveness. This, he said, provides a nice resource for experienced practitioners in the area, but is hard for beginners to fully comprehend.
“We aimed to provide a clear roadmap, together with a few of the most representative works, to help newcomers quickly enter and navigate this vibrant field,” Tong said.
That the resulting paper earned this distinction from the IEEE journal on Big Data Mining and Analytics became a nice recognition for the co-authors. But, Tong said, “more importantly, it reiterates the power of long-term collaboration, as well as the importance of integrating research and teaching.”