The 2019 Conference on Decision and Game Theory for Security, also known as GameSec, was held in Stockholm, Sweden this year, and the research paper awards were presented in a unique venue: a room that once housed a nuclear reactor on the campus of the KTH Royal Institute of Technology. The cavernous underground space was lit by candlelight and filled with the voices of opera singers, and the hole in the center of the floor where the reactor once sat was filled with a radiant blue light.

It was in this room that Northeastern alumna Lisa Oakley (BS, Computer Science/Mathematics, ‘18) received the award for Outstanding Student Paper for her paper “QFlip: An Adaptive Reinforcement Learning Strategy for the FlipIt Security Game,” co-authored with Professor Alina Oprea.

Oakley, who hails from Napa, California, graduated from Northeastern with a combined major in CS and mathematics. She currently works as a research assistant in Northeastern’s Cybersecurity and Privacy Institute. Oakley got her start at the Institute while she was still an undergrad, after being inspired by Professor Alina Oprea’s course in cryptography. In the course, students study problems related to computational information protection and cybersecurity.

“I took Alina’s cryptography class in the Spring 2017 semester, and I really enjoyed it and started looking up what she did,” Oakley says. “I saw she was working in the Cybersecurity and Privacy Institute on some machine learning research, so I went to her office hours and found out she works with undergraduate researchers.”

After leaving campus for her second co-op with a cloud-computing company, where she was able to gain some additional real-world experience watching cybersecurity in action, Oakley waited one more semester before reaching back out to Professor Oprea and beginning her research project.

Once affiliated with Professor Oprea’s lab, Oakley focused her research project on applying game theory to cybersecurity — specifically, working with FlipIt, a security game created in 2012 by a team at MIT and RSA Labs, where Oprea worked at the time, that models attacker-defender interactions to help understand potential real-life cybersecurity scenarios. Oakley first read up on past work with FlipIt, and Professor Oprea encouraged her to find a way to implement a reinforcement learning strategy for the game. Reinforcement learning is a machine learning method that uses rewards to reinforce positive choices made by the computer. Oakley also conducted theoretical and experimental analysis and set up her own custom environment for testing.

“Lisa worked on cybersecurity games for modeling advanced cyber attacks using techniques from game theory and AI,” Oprea explains. “Interestingly, reinforcement learning algorithms used for developing AlphaGo and other successful game agents can be adapted for optimizing defensive strategies in cybersecurity games. The potential in designing AI-based agents for protecting against advanced attacks is huge, and Lisa’s paper is opening up new avenues of research in this direction”.

Professor Oprea was the one to suggest that Oakley submit her research to the GameSec conference. Says Oakley, “Alina had this conference in mind because it really fits right in line with the work that we do. We worked together to compile all of our work into a paper, and we submitted it in June.”

At GameSec 19, Lisa receives the Outstanding Student Paper Award. L to R: John S. Baras (U of Maryland), Lisa Oakley, Tansu Alpcan (U of Melbourne), and György Dán (KTH). Not pictured: Eugene Vorobeychik (Washington U in St. Louis)
This was Oakley’s first time submitting an academic paper to a conference.

“I was very proud of the work that we did on the paper, we worked very hard on it and made it, in my opinion, a well-written and well-motivated paper,” Oakley says. “You can be happy with what you did [for a paper] and still not know if it’s going to be accepted, so I was really happy when I found out the paper had been accepted to the conference.”

GameSec 2019 was Oakley’s first visit to Sweden, and she says she gained a lot of unique insights from the experience. She is currently applying to PhD programs in cybersecurity and machine learning theory, and she says that “it was nice to get exposure to a lot of universities and professors from all over while applying to PhD programs.”

Oakley adds that cybersecurity was a natural fit for her because of her love of mathematics.

“I was interested in taking a course in cryptography because it’s an area of computer science with a lot of math in it,” she continues. “I was still trying to find the area of computer science I was most interested in. A lot of what we did in the class was model cybersecurity scenarios, because that’s what motivated the development of cryptography over recent history.” Personally, she says, “I found the scenarios all very interesting and compelling, and I liked that it had the mathematical element to it.”

However, there is also a practical facet to Oakley’s interest in the field. She explains, “Cybersecurity in learning theory feels very relevant and important right now, and it feels like a meaningful direction to go in, especially considering how quickly machine learning is being adopted to solve real world problems. You hear about all these different attacks that happen and data breaches… it just seemed like a worthwhile direction.”