“I was very quiet,” he says, “but I insisted on doing everything myself. Unlike my brother, who refused to wear his seatbelt, I was more than willing, but I absolutely had to buckle it myself, at age three. Those are good traits for a theoretician, or for doing research in small groups. But I really reject the stereotype of a lone-wolf approach to theory. I describe myself as a theoretician who’s ready and eager to work with people who are more practically inclined.”
Ellis is the most recent hire in the multi-year CS With Impact campaign, the largest expansion in Brown CS history. He recently earned his doctorate from Carnegie Mellon University, and after he completes postdoctoral work at ETH Zürich, he’ll return to the department in the autumn of 2023 as assistant professor.
Hershkowitz first came to Brown as an undergraduate, he says, because he was looking for the scientific and academic freedom to explore what he thought was interesting. Computer science didn’t appear on the scene until his sophomore year: “I took CS 17 on a whim and rather enjoyed it. Especially the theory, the mathy stuff.” Initially interested in neuroscience, then computational neuroscience, then CS, a “slow inching toward theory” paralleled his movement toward pure computation.
Taking an AI course with Brown CS Professor Stefanie Tellex and then joining her research group offered more math, more algorithms, and an exposure to machine learning. “That was what crystallized things for me,” Ellis tells us. “I knew I wanted more of that in grad school.”
Though wary of overstating his credentials as a philosopher, Hershkowitz sees his joint concentration in philosophy at Brown as significant. “I think of it as what ties together my CS research,” he says. “I really enjoyed twentieth-century analytic philosophy, looking at fuzzy, intuitive things in language and providing formalism to topics we care about. Being rigorous and formal is where I’m happiest, creating solid frameworks to look at things in different ways, asking questions and answering them in a specific and rigorous way.”
And he’s not inclined to see theory and practice as a binarism: “At one of the big tech companies, you might be working on a very specific problem and you need algorithms to solve it. You might not immediately implement the algorithms you’ve created, but the insights you gain are useful. Not every algorithm has to run.”
Ellis explains that most of his research revolves around graph algorithms, and he chuckles for a moment at their ubiquity as favorite objects of study in computation: “Dots and lines, nodes and edges are everywhere. ‘Road network, protein network? Looks like a graph!’ A lot of my work deals with asking how you can take an arbitrary graph and strip away unimportant structure, profoundly simplifying it while preserving what you care about. When a general model boils down to something simpler, the insights you gain are really useful. You can expect your algorithms to run more quickly, things to be computationally easier.”
Along these lines, some of his favorite recent projects with co-authors focus on ways to simply represent distances between nodes in potentially complex graphs. This includes new algorithms for removing vertices in a graph while preserving the distances between the remaining nodes and ways of representing distances in arbitrary graphs by simple graphs: namely, trees. By simplifying the networks at hand, such methods facilitate new algorithms for graphs, leading to efficient algorithms for designing networks that are robust to communication failures and conducive to low latency communication.
“I really think,” Ellis says, “that there’s a strong aesthetic component to my work, a ‘let’s try and peer into the heart of the universe’ feel. There’s a sense of building something, of putting together pieces: ‘I combined all my lemmas, and everything else follows.’ Abstract math can feel like woodworking sometimes, which I like.”
Sharing that aesthetic satisfaction and helping students build the skills needed to achieve it are what Hershkowitz is looking forward to most with his return to Brown: “It’s the teaching, getting people excited about what they’re learning. As an institution, Brown is genuinely interested in eclecticism, which allows for an incredible number of ‘CS plus seemingly totally unrelated other thing’ research projects. Our students are constantly learning about areas that they’d never encounter at some other school.”
“And I think theory has a real place at the table with our socially responsible computing initiative,” says Ellis. “Computer science has become such a well of power, and the idea of making it available to everyone excites me. Theory lets us look at questions of fairness, formalizing which outcomes are fair, and making those conversations rigorous. That’s a really valuable contribution.”
Brown CS alums often have valuable insights for today and tomorrow’s students, so we ask Ellis what advice he might have for them: “I’d tell undergraduates to get involved in research, just like I did. Brown CS is very unusual in that it combines top-tier research and strong grad students while really caring about undergraduate pedagogy. Among other things, that creates an on-ramp to research for undergrads in a very special way.”
His advisors at Brown CS really supported him, Hershkowitz says, and turned research into something that was possible and that he could really enjoy: “Stefanie got me doing research as a junior, and I wouldn’t have ended up in academia without her. Michael Littman and George Konidaris were tremendous mentors and huge influences on me. They all share a really genuine interest in working with undergrads.”
As we close the interview, Ellis decides to add a slight modification to the research pitch that he’d make to undergrads. Once again, he’s looking to share one of his great loves.
“When I was here at Brown,” he says, “someone put up these memefied posters around campus that said, ‘It’s never too late to do CS’ with a picture of a celebrity doing something. I’d like to make ones that say, ‘It’s never too late to do theory.’ I always thought you had to be a Math Olympiad person who was doing college-level math at age six in order to be a theorist, but it’s not true. People think they don’t have the background, but I want to be the one to tell them that they can totally do it.”
For more information, click the link that follows to contact Brown CS Communications Manager Jesse C. Polhemus.