Surbhi Goel is an assistant professor at the University of Pennsylvania, where she joined in 2023 after a postdoctoral fellowship at Microsoft Research New York. She completed her PhD at UT Austin Computer Science (UTCS) in 2020, working at the intersection of theory and machine learning. This year, she was awarded the prestigious Alfred P. Sloan Fellowship in Computer Science: one of only 22 early-career computer scientists selected across the US and Canada. We sat down with Surbhi to talk about that recognition, her time at UTCS, and what she'd tell students charting their own research paths.
You recently received the Sloan Research Fellowship for Computer Science. What does that recognition mean to you?
What makes the Sloan so meaningful to me is what it actually recognizes. It isn't tied to a single paper. It's an investment in you as a researcher and in the direction your work is taking.
My work sits at the intersection of theoretical computer science and machine learning, and a lot of it is about making modern AI systems trustworthy, understanding why models succeed and fail, and building algorithms that come with provable guarantees. As these systems get used more widely across society, that kind of foundational understanding becomes more important, not less.
I never expected to be selected, and many of the people I look up to have received this same fellowship. Being acknowledged alongside them feels like a sign that the community I care about genuinely values what I'm doing.
Looking back at your time at UTCS, what moments or opportunities were most central to your growth as a researcher?
A few moments really stand out. The first was starting to work with my advisor, Adam Klivans, in 2016, just as he came back from a sabbatical. The timing was perfect, because he was ready to dive into something new. Deep learning was still relatively new as a research area. We'd seen striking practical success starting around 2012, but the theory of deep learning was nascent, and not many people were working on it. Adam and I had a nice conversation and realized we were both excited about this, and the gap between what worked in practice and what we could explain was wide open. We ended up pursuing that direction for the rest of my PhD, and I'm still working on it.
What made Adam so important wasn't just the research direction, but that he trusted my abilities before I did. In the early stages of a PhD, you don't have much confidence in your ability to do research. It's hard, and it takes a while to see any success. It mattered a lot that Adam trusted I would get there. He also encouraged me to work with other people and explore my own research interests, which shaped me as much as the technical work did.
The second was my cohort. UT is a nice community because it isn't very big. My incoming class was about 30 students, and we all took a class together that first semester and became good friends. Over the years, that group became one of the most important parts of my PhD. They weren't only a support circle. They pushed me, challenged my ideas, and talked through hard problems with me, and they gave me perspectives from outside my own research area that shaped the kinds of questions I chose to work on. A lot of them are still my closest friends!
The third was reaching across departments. At the time, very few people at UTCS were working on theoretical machine learning, so in my first year I reached out to Alex Dimakis in UT ECE and asked if I could sit in on his group meetings. He was incredibly generous and welcomed me right away. I attended his group meetings throughout my PhD. I was never formally part of his group, but he always made me feel welcome, and I ended up collaborating with people there too. Those cross-department connections were invaluable.
And toward the end of my PhD, two visits were pivotal. I spent a summer at the Simons Institute at Berkeley as a research fellow, during their program on the theory of deep learning, and later I visited the Institute for Advanced Study in Princeton. Both put me in a room with the broader theoretical machine learning community. I met peers from other universities, got to talk about my work, and many of those conversations turned into collaborations and invitations that lasted for years. They really widened my sense of what the field was and what was possible.
What advice would you give PhD students, especially those working in a niche or emerging area with less existing research to draw from?
Lean into it. Working in an emerging area is exciting precisely because you get to decide what problems people care about in that space. You're defining the first questions worth asking, and getting to shape a space that early is a rare and exciting position to be in.
Then talk to people, constantly. Share your ideas as you develop them so you can get feedback early, and reach out beyond your own group, not just at UT but elsewhere too. The visits I mentioned earlier were valuable for exactly this reason, but you don't have to travel to do it. UT brings in a lot of speakers, so go meet them, talk to them, and be open-minded. You'll surface problems you wouldn't have thought of on your own. Talking to people has been my biggest source of problems, solutions, and motivation, and it's something you can do at any stage of your career.
And seek out opportunities rather than waiting for them to appear. There are far more available than you might think. I hesitated in my first few years and wasn't reaching out, but once I did, it multiplied the reach and impact of my work in ways I hadn't anticipated.
What would you say to someone considering a PhD at UTCS?
The biggest thing is the community. UT is warm, welcoming, and genuinely homey, and that kind of environment matters more than people sometimes realize. A PhD is a long journey, a marathon rather than a sprint, and you need a lot of support to reach the end. A community that carries you through is what makes it possible.
Beyond that, UT has tremendous resources. There's significant compute available if you want to run big experiments, which is rare in academia and a real differentiator for UTCS. A large number of faculty have joined in the last five or six years, and the NSF AI Institute for Foundations of Machine Learning has brought in even more researchers in theoretical machine learning, along with speaker series, workshops, and events across the university.
And then there's Austin itself, which is one of the best cities I've spent time in. The sunny days really do make a difference in how enjoyable the PhD experience is.
Ready to write your own story?
UTCS's PhD and MS research programs are ranked among the top 10 in the country, with deep expertise across AI, machine learning, theory, and beyond. Take advantage of our compute resources, cross-department connections, and mentorship to help you do your best work.