By Shandana Mufti

When Sarah Brown receives her PhD in electrical engineering this December, she will be third in a line of PhD holders convened at Northeastern, and an “academic granddaughter” to CCIS Dean Carla Brodley.

“Sarah just really amazingly stood out in all ways, both for what she was doing in terms of her research and then also some of the other extracurricular activities that she’s done,” Brodley says. “I’m really proud of Jennifer and of what Jennifer has achieved with Sarah, and of Sarah herself.“

Brodley was advisor to Jennifer Dy, who earned her PhD at Purdue University in 2001 and is now a professor in the College of Engineering’s Department of Electrical and Computer Engineering. Dy is advisor to Brown, who will defend her thesis, titled “Machine Learning for Computational Psychology,” in the fall. Ever since Brodley joined CCIS in 2014, Brown says she’s seen a resemblance between Dy and Brodley and how they interact with the students they mentor.

“It’s clear that they’re very intentional about how they advise,” Brown says of Dy and Brodley. “It’s funny, now that [Brodley’s] here, it’s interesting to see how similar she is to [Dy] and what things they do that are similar.”

Brodley says that by figuring out ways to guide her students to move past problems themselves, she mentors students to be independent thinkers who will be empowered to lead research themselves in the future. She recently received the Harrold and Notkin Research and Graduate Mentoring Award from the National Center for Women and Information Technology, an award named for pioneering computer science researchers and mentors. Brodley’s mentoring style has been a source of inspiration for Dy as she guides her own Ph.D. students through their research. That’s meant listening to students and adapting to match each one’s strengths so that they can succeed. “Let them grow,” Dy says. “I think that’s key.”

And Brown has excelled during her decade at Northeastern, first as an undergraduate student, and now as a Ph.D. candidate. She entered her graduate studies with fellowships from Draper Laboratory and the National Science Foundation, and has spent the past five years tackling several projects related to machine learning.

“They share the common thread that we want to study these problems in the brain,” Brown says, explaining how her three projects tie together. “The tools we have don’t allow us to answer these questions, so whenever we get results, we need a way of knowing whether they’re good. Then we can look at case studies of PTSD and affect.”

The first piece of her work looks at stability performance measures that test algorithms by removing small chunks of data from a large data set and running the model, then repeating the test with a different bit of data removed. The solution should be very close each time – if not, the algorithm isn’t working. Stability can be used to assess how well other performance measures work. “These are intuitive, understandable performance measures but we want to make sure that we understand how much they actually mean and what types of quantitative statements we can make about them,” Brown says.

The second part of her work looks at diagnosing post-traumatic stress disorder, or PTSD, from peripheral physiology like heart rate and skin conductance. After undergoing structured interviews, subjects were given a score from 0-140, where 0 is no symptoms. Then, once the data was collected about each subject’s peripheral physiology, a scoring function was used to predict a subject’s score.

The final part uses functional MRI (fMRI) scans to analyze how the brain responds to pictures shown to subjects, and then use machine-learning algorithms to find patterns in that data. Usually, trials like these show participants about 30 pictures and look for one response shape to light up in specific places for each image type – Brown’s collaborators from the Interdisciplinary Affective Science Lab showed a group of five subjects 900 pictures over several visits. “We’re going to repeat standard fMRI analyses and show if we do it regularly, this is what we get, same results,” she says. “If we add more and more data, more and more of the brain’s engagement is visible.” With the data set from the study, Brown can use the brain responses to figure out which images were similar to the subject, and review them to determine what was similar about the images.

After Brown completes her Ph.D., she’ll move to the University of California Berkeley as a Chancellor’s Postdoctoral Fellow, where she’ll continue her work under Michael Jordan, a leading machine learning and artificial intelligence researcher.

Dy says she’s proud of her student and of the work she’s completed, and she’s outspoken about her hopes that Brown will also join the world of academia. “She’s on the right track,” Dy says. “Hopefully, she’ll become a third generation faculty member.”