6/15/2022 10:55:00 AM
From Josep Torrellas partnering on a project with Illinois CS collaborators for Intel’s Center on Transformative Server Architectures to Geoffrey Herman’s NSF Award for Cybersecurity Instruction, four newly funded projects will push computing forward.
Having long-established himself as an impactful researcher in the art of server computer architectures, Illinois Computer Science professor Josep Torrellas was thrilled to learn about Intel's recently issued Call for Proposals to create the Center on Transformative Server Architectures.
Together with several of his colleagues, Torrellas was already prepared to investigate several subjects that would be relevant to this new center’s initiatives. These subjects included virtual address translation, processing near memory, heterogeneous memory systems, and networking support.
The topic he proposed with collaborators Nam Sung Kim, professor of Electrical and Computer Engineering, as well as Computer Science professor Tianyin Xu received funding and is entitled “2030 Server Architecture for Terabyte-Scale Heterogeneous Computing and Memory.”
“Professor Kim from ECE is one of the leading experts in memory systems design and architecture. Professor Xu is a leader in how operating systems interact with computer architecture. I have been working with them for multiple years, and I know their great intelligence and expertise,” Torrellas said. “I complement them with expertise in processor and memory systems architecture. It was natural for me to apply with them.”
The three-year award from Intel will provide $200,000 per year.
Torrellas said that he, Kim, and Xu have the goal in mind of making server architectures “10 times more performant and 10 times more energy efficient” than they are today. Their belief going into the project is that this can be done by making key changes to the hardware of processors, memories, and network interfaces.
“We hope to revamp the virtual address translation in processors, devise near-memory processors, develop new ways to ensure processors, GPUs, and accelerators operate in a tightly coupled manner, and improve network interfaces. I hope some of these designs will improve Intel products,” Torrellas said.
He also complimented the major investment Intel provided, as the concept of the Research Center provides a new opportunity for multiple research groups from around the world to come together on this topic.
“It is a unique opportunity to make a positive impact on the company,” Torrellas said.
Adam Bates Earns Three-Year VMware Funding
Illinois CS professor Adam Bates’ research group, The Secure & Transparent Systems Laboratory, has dedicated itself for some time to the concept of data provenance.
Their belief is that the topic can “revolutionize” how we defend systems. In April, VMware expressed belief in their effort to produce on this topic, as it provided a three-year, $225,000 gift in support of the team’s efforts.
“Data provenance describes immutable truths about intrusions that are impossible for attacks to evade or conceal; all we need to do is ask the right questions of it,” Bates said. “Partnering with VMware and Carbon Black will give our research group an opportunity to generate compelling evidence in support of that vision.”
The way data provenance works, according to Bates, is “as a special form of auditing that, rather than examine the system as a stream of discrete events, parses those events into a casual dependency graph that describes the history of system execution. It thus allows us to reason about suspicion activities through their historical context within the system.”
Bates said that his group has conducted one-off collaborations with industry in the past, helping them to discover the tremendous potential of data provenance to improve threat detection and monitoring products.
“For example, even state-of-the-art threat detection products like Carbon Black can be prone to false starts,” Bates said. “We’ve discovered that provenance is very effective in triaging alerts to prioritize those that are likely legitimate attacks. This tells us that seemingly-covert intrusions are actually fairly obvious when we reason about them through the lens of their provenance.”
Bates and his team look forward to continuing their regular meetings on this topic with David Ott, of VMware Research; Matt Lentz and Raghava Batta, of VMware Faculty Affiliates; and Christophe Briguet of Carbon Black.
But he also credited his PhD advisee, Pubali Datta, for laying the groundwork for the project by “advocating for provenance-based solutions” during her VMware internship.
Looking forward, Bates is eager to investigate a series of questions that he hopes can turn into active collaborations with VMware and Carbon Black.
“My research group is also investigating a variety of other questions related to data provenance. Can provenance improve host-layer anomaly detection? How do we efficiently store and analyze provenance? How can we use data provenance to more effectively explain threat incidents to analysts? And so on,” Bates said. “The most valuable resource VMware has to offer us is their practical perspective and experience, which ensures that our academic research is grounded and impactful.”
Talia Ringer Wins Amazon Research Award with Cross-Institutional Collaboration
Talia Ringer joined Illinois CS faculty last year with a dedicated research interest in programming languages/formal methods/software engineering to prove the absence of costly or dangerous bugs in software systems.
Until recently, that work did not intersect with Machine Learning (ML). But in a project that began with traditional programming languages techniques, Ringer found it to be a great benefit. When a problem with generating human-friendly abstractions sprang up, the process veered toward finding ways in which ML could help.
However, they also knew from past experiences that work with ML struggled to find semantic relationships between datatypes and struggled to manipulate the low-level proof objects. Meanwhile, traditional tools are quite good at this.
“So, I got a bit obsessed with how to bring together the two approaches,” Ringer said.
In support of this goal, they received a one-year unrestricted Amazon Research Award for the project, which is now entitled “Neurosymbolic Proof Synthesis and Repair.”
The cross-institutional collaboration includes researchers from the University of Massachusetts Amherst.
“The grant is looking at bridging the gap between traditional programming languages techniques and machine learning for automatically 1) generating formal proofs of program correctness (proof synthesis) and 2) fixing formal proofs of program correctness in response to changes in the programs or in the specification of correctness (proof repair),” Ringer said. “Machine learning has shown a lot of promise in helping improve proof automation in recent years, but still struggles to make use of semantic information. So, we are looking at different ways to train a model to pick up on that semantic information and use it to improve model performance for both proof synthesis and proof repair.”
They also have plans to take ML forward in a few other projects, too, as the research indicates a new road to explore with students willing to help push the boundaries.
“I’m especially excited to keep exploring this in parallel to my exploration of improving the state of the art in traditional proof automation,” Rigner said. “For me, traditional proof automation is like carving a path to advance the state of the art and go somewhere nobody has ever gone before – and machine learning has the potential to pave roads over those paths and bring proofs to the masses.
“It's fun for me to lead a group of students that's unafraid of exploring both type theory and language models and everything in between.”
Lingming Zhang earns 2022 Google Research Scholar Award
Finishing his second academic year with Illinois CS, Lingming Zhang considered it quite an honor to learn he had earned inclusion as a 2022 Google Research Scholar Award winner.
Designed to support early-career professors who are pursuing research in fields relevant to Google, Zhang’s work in Software Engineering and Programming Languages earned him this particular award. Specifically, the program is for a project Zhang entitled “Towards Effective Fuzzing of Deep Learning Libraries.”
“I am very proud that Google Research felt highly enough of the potential for this research effort to include me as part of the 2022 Google Research Award winners,” Zhang said. “Since joining Illinois CS, I have been supported by my peers – who are too many to name here – and I have always been inspired by the Illinois students. I would say that this award would not be possible without the hard work of all my excellent students and interns in the Illinois iSE group. This is another step toward furthering our research efforts in fuzzing deep learning systems, as well as our overarching goal of synergizing software engineering and machine learning.”
Geoffrey Herman’s NSF Award Focuses on Teaching Cybersecurity Effectively for All Students
As a teaching professor researching Computers and Education, Geoffrey Herman’s recent NSF Award ties to cybersecurity work he began in collaboration more than five years ago.
Together with collaborators Alan Sherman and Linda Oliva at the University of Maryland, Baltimore (UMBC), Herman began the Cybersecurity Concept Inventory. He is now excited to start using the Cybersecurity Concept Inventory to more rigorously study whether different teaching methods are more effective at helping students learn cybersecurity concepts.
Their work will now continue through a three-year NSF Award totaling about $500,000 across three institutions in support of a project entitled “Examining Pedagogy in Cybersecurity at Military Academies.” UMBC is the lead institution, with the University of Illinois Urbana-Champaign and the University of Minnesota Duluth involved.
With Sherman’s UMBC campus located in close proximity to the US Naval Academy, Herman said that Sherman cultivated a few relationships with students and colleagues who had close affiliations with the Naval Academy.
“One of the really interesting things about the military academies is that several of them teach cybersecurity to all of their cadets, not just those that are interested in computer science,” Herman said. “As you might imagine, it's vital for every member of the military to be concerned about every form of security, including cybersecurity. Consequently, we are super excited to team up with the military academies, specifically, because their cybersecurity courses must be inclusive of students with any academic interest.
“That, in turn, lets us really begin to explore the question: How do we teach cybersecurity effectively for all students?"
The workgroup can then place its attention on the primary focus of the project: To study whether simulation-based based pedagogies actually help students learn cybersecurity concepts and how they can package those simulations into easily reusable packages for other instructors to use.
One example the group could use is to give some students the task of trying to schedule a meeting through an instant messenger platform while other students are tasked with being interlopers who can intercept, change, or block some messages from getting through with the goal of creating mayhem.
“This activity might teach students to realize that many cybersecurity attackers are not just concerned with stealing information but have broader nefarious goals such as simply disrupting how an organization functions,” Herman said.
The group will work with more than 30 instructors from military academies who teach cybersecurity.
Considering that the Illinois CS professor’s academic passion is to help student learn more while also finding scalable, sustainable ways to help faculty teach better – Herman considers this project to be a marriage of both goals.
"We hope to work closely with these instructors to understand the challenges and successes that they experiences as they adopt and adapt these simulations in their classrooms,” Herman said.