Graduate student profile: Yaniv Yacoby

Yaniv Yacoby faced a choice early in his Ph.D. candidacy. His advisor at the time, Margo Seltzer, left the Harvard John A. Paulson School of Engineering and Applied Sciences (SEAS) for a faculty position at the University of British Columbia, but Yacoby didn’t want to be that far away from his family in Massachusetts. Instead, he switched to a new research group at SEAS,  the Data to Actionable Knowledge (DTAK) Lab run by Finale Doshi-Velez, Gordon McKay Professor of Computer Science.

“I came in with no math or statistics background, which is pretty much all of what I do now, so the transition was very difficult,” Yacoby said.

Yacoby will receive his Ph.D. in computer science later this month, but still remembers the challenges of such a drastic academic shift. That was a motivation for devoting as much energy to promoting mental health and student wellbeing on campus as he did to the lab and his studies. He co-created and taught “CS290: Effective Research Practices & Academic Culture,” has represented SEAS on the Harvard Student Wellbeing Council, and has been a peer mentor with the InTouch graduate student support group.

“If you look at the statistics about mental health in Ph.D. programs, it’s not just Harvard that has very high rates of anxiety, depression, isolation and impostor phenomenon,” Yacoby said. “I was definitely falling into that bucket, and it led me to reflect on the ways we can create a better academic culture. If we’re not careful, all of us, students, faculty and administrators, we unintentionally fall into habits of perpetuating a noninclusive and toxic culture in academia. It’s very important to me that people succeed here and leave behind this idea that only some people are good enough to do a Ph.D.”

As much as Yacoby wants to improve the academic culture everywhere, he’s also tried to model inclusion and supportiveness in his own lab. Doshi-Velez has seen it firsthand.

“He's always looking out for other students in the lab,” Doshi-Velez said. “He will cheer for them, provide technical advice, and also that wisdom of how they can grow individually and how they can ask for support.”

DTAK focuses on developing machine learning methodology to address many decision-making scenarios involving humans and artificial intelligence (AI), especially in healthcare, as well as broader social and technical questions around human-AI interaction, AI accountability, and responsible and effective AI regulation. Within DTAK, Yacoby’s research focuses on deep Bayesian models, a class of methods that recasts new advances in deep learning like neural networks into a statistical modeling framework. This framework provides a principled method to quantify uncertainty, which is crucial in safety-critical applications.

“Let’s say you’re given information about a patient, like their body mass index, age, etc. and you want to know what will happen to their blood pressure after you give them a certain intervention,” he said. “It’s not enough to just make a prediction about what will happen – you also want to know how uncertain the model is and why. Is the model uncertain because it hasn’t seen similar patients, or because it’s missing some patient attribute that interfered with the intervention? A lot of my work focuses on developing methods that capture these different types of uncertainty, while still being able to model complex data. This is challenging because it requires solving problems that even modern computers are quite slow at solving. But figuring out how to get around this problem is something I really enjoy.”

Doshi-Velez remembers being immediately impressed with Yacoby in their early meetings when he joined DTAK. That admiration has only grown as Yacoby has delved into his Ph.D. research, which she described as “very subtle but important issues that come up with training highly-complex machine learning models.”

“What impressed me about Yaniv was that he was not only a highly-independent learner but also a great communicator,” she said. “We would have very efficient meetings where he'd summarize what he had learned or was thinking about, ask questions, and we could sketch out a plan for the following week -- and then he'd come through with consistent progress.”

Yacoby first arrived at Harvard as an undergraduate, enrolling in the Dual Degree Music Program. He earned an A.B. in 2015 in computer science, and a year later finished his master’s degree in contemporary improvisation from the New England Conservatory of Music. As a senior, Yacoby combined his two interests by designing real-time audio processing tools that allow performers to augment and transform the sounds of their instruments.

“I took Professor Hans Tutschku’s course, which taught me a new way of listening to music as raw sounds, rather than letting your brain translate it into rhythm, melody, harmony – the standard components of music coming from the Western classical tradition,” he said. “I got really interested in what you could do with raw sounds, so I came up with a bunch of tools where you play the instrument the way you normally would, be expressive the way you normally would, but it would, in real time, translate the sound into something else and adapt along with you.”

After receiving his degree, Yacoby will do a one-year postdoc in the Nock Lab at Harvard, where he’ll develop new machine learning methodology to help ongoing efforts to understand the psychology underlying suicidal ideation. After that, he’ll join the Wellesley College faculty as assistant professor computer science.

“I’m excited to be teaching students, and learning from them,” Yacoby said. “I want to use this dialogue with students as a way to continue exploring the broader impacts of machine learning systems – how they should and shouldn’t be used – and to continue pursuing my research program and incorporating students into my projects. I’m particularly excited to join Wellesley because it’s such a great fit for all my interests.”

Yacoby was willing to try something totally new early in his Ph.D. candidacy, and it worked out quite well. And thanks to a positive experience in his lab and at Harvard, he’ll soon get to bring his research and philosophy to a whole new group of students.

“I’m excited to continue thinking about how we can use curricular and extracurricular interventions to shape academic culture, be more supportive and inclusive, and empower students to prioritize community and mental health,” he said. “Finale’s lab has an incredibly supportive and thoughtful culture. I definitely would not be here if not for them.”