Three Illinois Computer Science Faculty Earn NSF CAREER Awards

5/15/2018 10:42:20 AM David Mercer, Illinois Computer Science

Three more Illinois Computer Science professors have received prestigious National Science Foundation CAREER awards this spring.

The NSF announced the most recent awards to Assistant Professor Ranjitha Kumar to advance her work on mobile design, Assistant Professor Matus Jan Telgarsky for a project designed to create a theoretical understanding of the foundations of neural networks, and Assistant Professor Adam Bates to finance research on using data provenance as a security monitoring tool in distributed systems.

Those awards follow Assistant Professor Ruta Mehta’s CAREER Award earlier this year.

CAREER Awards are given by the NSF’s Faculty Early Career Development Program and recognize junior faculty who have the potential to serve as academic role models in research and education.

A look at the awards given to Bates, Kumar, and Telgarsky:


Security was once focused entirely on keeping attackers out of computer systems, but in the age of cyber warfare, Assistant Professor Adam Bates says this is changing -- if a powerful attacker wants to break into your system, eventually they will. What really matters is what happens next.
Assistant Professor Adam Bates.
Assistant Professor Adam Bates.

“Particularly when we start thinking about these incredibly well funded nation-state attackers,” he said, pointing to a series of attacks in 2009 on Google and other companies known as Operation Aurora. Attackers searched for months for the email accounts of human rights activists before the companies could lock the attackers out.

Bates plans to use his NSF CAREER Award to advance the use of data provenance (a detailed history of system events) to detect attackers who break into networks, figure out what they’re up to, and -- as quickly as possible -- prevent them from achieving their objectives.

Bates wants to use binary analysis to unify the information stored in individual application logs to help detect anomalies across a distributed system.

One key problem is how to scale this “rich, fine-grain record” of log information gathered across dozens, hundreds, or thousands of machines that could be compromised.

“We eventually have too much information, right? And so how can we efficiently distill that into something that's an actionable piece of information for a security team or a system administrator?” Bates said.

A lot of the activities across the many computers in a distributed system are repetitive. Bates plans to apply grammar learning mechanisms to identify the unusual or unique activities at each particular node in order to peel away repetitive information, then distill what’s left into a relatively simple, unified graph that a system administrator can use to act in almost real time.

“If we’re incrementally building these relationship graphs, you can actually query them on the order of microseconds,” Bates said.

Bates’ work on data provenance has attracted interest from Visa, Wal-Mart and MIT Lincoln Laboratory. He hopes his work will lead to new ways of protecting data and critical infrastructure.

He also in the process of building a new security research group, the Secure & Transparent Systems Laboratory, and plans to work with two of his PhD students on the NSF-funded project.


Assistant Professor Ranjitha Kumar plans to build on her longstanding work in mobile design, capturing and aggregating design features and user interactions from a wide range of existing mobile apps and using them to develop a platform where designers can access them.

Given the time people spend using mobile phones and other devices, digital design is ever-present, and large amounts of data are constantly being generated about how users interact with that design. But the
Assistant Professor Ranjitha Kumar
Assistant Professor Ranjitha Kumar
companies behind those designs seldom share data about how they are used.

“You’re not taking a step back and saying, ‘Of all the possible ways one could do this, are we doing it the best way?’” Kumar said. “My goal with this proposal is to break down these analytic silos, and open-source usage data so people can understand ‘this is the best way to do something.’”

Kumar will use the award to clear technical hurdles.

First, she plans to develop scalable, privacy-preserving systems that capture design and interaction data as people use apps on their phones. This work will build on her existing ZIPT platform, which allows designers and marketers to run targeted usability and design-performance tests at scale over apps they do not own and did not build.

The Rico dataset of app design she built will also be put to work on the project.

“The key insight here is that, if we can use understand the semantic ‘building blocks’ of design, we can start to aggregate user interaction data,” Kumar said. “We can begin to identify the UX patterns and design choices that truly lead to better user experiences.”  

Ultimately, she hopes to provide designers with tools that will allow them to find the best possible design solutions and the evidence they need to argue in favor of designing better apps.

“So much design focuses on reinventing the wheel. The purpose of my group’s work is to save you from spending your own time and engineering resources on problems that other people have already solved,” she said.


While neural networks have become central to the performance of artificial intelligence, Assistant Professor Matus Telgarsky’s plans for his CAREER award are driven by what isn’t known about them, including in some instances why and how they work.

“Neural nets have parameters that fit to data. The procedure which you use to fit to the data is, essentially, more than half a century old at this point. No one knows why it works for neural nets,” Telgarsky said. “So part of the proposal is asking three basic questions of why this thing works.”

First among them is representation, what facts or events can a neural net reasonably approximate? Second is optimization, how to best fit
Assistant Professor Matus Telgarsky
Assistant Professor Matus Telgarsky
neural networks to data. Telgarsky brings deep expertise in optimization to the project.

The third question, he said, could prove to be the most challenging – generalization, the question of how neural networks fit not only data that they have seen, but how they are likely to handle data they haven’t yet been exposed to.

If, for instance, an autonomous vehicle relies on a neural network, that network needs to be able to process circumstances that it may not yet have seen.

“You need to have some kind of sense that it can work well on future examples,” Telgarsky said. “Maybe it only saw pedestrians in sunny conditions. You want to make sure it still works in other conditions.”

To deal with that third question, Telgarsky said, will likely require what he describes as a long, tedious process.

“Just to basically stare at neural nets running in the wild a lot -- I literally just stare at the plots of what they do,” searching for patterns, he said.

Telgarsky also stresses that the award means that he has the ability to pay and retain his students for the research.

“My students can spend full time on research problems,” he said. “I really want to take care of my students.”