Researcher of the Month
Majors: Computer Science (Honors program), Math; Minor: Political Science, Class of 2021
Research Mentor: Dr. Robert Kelly, Computer Science
Gilvir Gill is a junior double-majoring in Computer Science (honors program) and Mathematics with a minor in Political Science. And from the get-go, joining the VIP-PoliTech team was the clear path for exploring his interests in a meaningful way. Gilvir reflects: “It's amazing that there's so much material here that I'm able to still be excited about something after working on it for a year and a half!”
Is there a fair way to draw electoral district boundaries? What are the challenges to automating this process? The VIP-PoliTech team, under the mentorship of Prof. Robert Kelly (Computer Science), has been working hard to develop and optimize an algorithm for the Automated Redistricting System, a system for randomly generating congressional district maps using graph partition techniques and Markov Chain Monte Carlo approaches. Gilvir has contributed by aggregating census and election data from various sources; coordinating three subteams to improve and visualize redistricting algorithms; and creating a frontend visualization as part of the PoliTech Automated Redistricting System to smoothly display tens of thousands of electoral precincts simultaneously in large states. Gilvir’s enthusiasm for the VIP PoliTech project, and his awareness of the project’s relevance, is unmistakeable; "In a couple of months, the Census data will be available to everyone and the commissions will start doing the next round of redistricting …. Knowing that every minute I put into this matters right now because theoretically, that's a little bit more information for the redistricting commissions to use. So that really helps me realize how important this work is."
On campus, Gilvir has also served as a Teaching Assistant for CSE / Honors Theory of computation. Long term, he aspires to pursue a Ph.D. in Computer Science. Gilvir is a graduate of Stuyvesant HS, and is a life-long resident of Queens, NY. His hobbies including baking, aspirational ukulele and keyboard playing, and reading Robert Caro, Steve Kornacki, and other political journalism/history writers. Below are excerpts of his interview with Karen Kernan, URECA director.
Karen: Tell me about your research.
Gilvir: The main thing I’ve been working on with Dr. Kelly and the rest of the PoliTech VIP team is looking at algorithmic approaches to redistricting. One of our goals is to create the best automated redistricting system that we can, in terms of creating results that would actually be usable by redistricting commissions—that is, trying to look for some sort of globally optimal solution to redistricting problem which we end up treating as a graph problem on a precinct level. And a second approach of ours is to generate as many districts from a random distribution as we can so that we can kind see how existing redistrictings fit into an overall distribution. For example, if there's suspicion that a state is exceptionally gerrymandered, we can compare it to the sample of 10,000 purely randomly generated redistrictings that meet all the compactness and population equality constraints that you'd expect from congressional districts. So these approaches are algorithmically very related.
This sounds like a challenging problem.
Yes, automated redistricting is still a really open problem. I started working on this, about a year and a half ago when I joined the PoliTech Team. But within the community of applied math and computer science people who have been working on these sorts of problems in recent years using computational methods to create random districtings, there's actually still a lot of work that could be done to improve the techniques. One of the main issues is that there are a lot of constraints that need to be met for an automated redistricting to be usable by a state.
For example, a lot of states expect that districts ought to be drawn somewhat along county lines. You don't want to split counties at just arbitrary points. There are also other constraints that needs to be considered involving racial proportionality, plus a lot of legal issues that also need to be considered.
I think what makes the PoliTech team so unique is it has to be truly interdisciplinary for us to work on these problems. A lot of the work that Professor Kelly does is communicating with people in the Political Science department or people that are involved in the legal community around redistricting so that we can better understand all of these restrictions, and then quantify them in a way that works nicely with solving optimization problems in the realm of mathematics or computer science.
How did you first get involved in VIP?
The VIP teams have a unified application process you can find on their website. I got an email about it at some point, and when I looked through the team list and description, and saw that it dealt with political gerrymandering, I knew I was interested right away. I'm a computer science and math major with a minor in political science. So it was a perfect fit for me because it combines all three of my interests into a single project. And so I joined the team, my first semester as a sophomore.
At the time, it was a pretty small team. That was really nice, because the senior students on the team who were about to graduate made an effort to get me on board and up to speed on everything so that the project would continue. And by only my second semester, after a lot of the other team members had graduated, I was the most senior member of our VIP project, and I had the job of helping train new members.
How many students do you have on your team now?
So we started with ~7 and last semester had roughly 25 or so students working on it.
Does your team have a mixture of majors?
We’ve had four or five political science people, and psychology majors too, as well as computer science majors. And they're all very important members of the team. Something that I find really cool about the team is that I get to work with people whose main disciplinary focus is completely different from mine. And I can honestly say that being a part of this VIP team has improved my experience at Stony Brook tremendously..
How has Professor Kelly shaped your experience of VIP?
Oh, he's amazing! I think what really kind of makes Professor Kelly stand out is that he's so willing to set aside time to talk to individual members of the team about what they're working on. I have one-on-one meetings with him, essentially every other week for an hour and a half or two hours. And he's so passionate about this project. All of us are.
Have you learned things that you wouldn't have learned in classes from your involvement?
Actually, the requisite knowledge you need is something that I'd say ~ 90% of CS sophomores and juniors have already learned through their classes—of which, spanning trees is probably the most important topic. Taking AMS 301 for example is almost enough knowledge to get you started immediately.
But I'd say that my involvement with VIP PoliTech helped me take my to consider my courses under a different light. The material I’m learning somehow seems a lot more useful now because when I am learning, I can think about how I can use this in our project. In thinking about comparing two random districtings, for example, there’s an issue that you'd run into such as; what if your districts numberings don't line up in the nicest possible way? …Then you remember a problem that we spent maybe 20 minutes on in CSE 260 discussing, which ends up being really useful for handling this issue. And so it makes you realize that all the things you're learning and your courses are incredibly useful and also helps you apply them in more meaningful ways.
Do you find it helpful to have a long term involvement with a project? Were you glad that you got started in sophomore year?
Yes, I think that's been really helpful for me. But I would also want to point out that it’s not necessary to have a lot of background to initially get involved. I'm as involved in the projectsnow, as I was eight months ago. And …well first of all, it's amazing that there's so much material here that I'm able to still be excited about something after working on it for a year and a half! We have a lot of new students coming on the team this semester. And I'm hoping that they'll have the same experience that I did when I started, and that they’ll be really just be able to jump in and start having fun working on this immediately.
In your experience with the project, what's been one of the most surprising things that you've learned ?
Political redistricting is such a big issue. But what a lot of our work has kind of shown Is that, unfortunately, a lot of the biases in redistricting can somewhat be explained by inherent geometric features of a state. You have this clustering of democratic populations in your cities, whereas rural areas tend to vote Republican. We and other groups working on this problem have looked at particular states such as North Carolina, comparing the enacted plans to a large random sampling, and shown that they are in fact more biased towards Republicans, for example, than they should be. As a computer science student, it's not my job to speculate on how that problem should be resolved. But I still find it really cool that it's something we're able to kind of show, by looking at thousands of random redistrictings.
What were some of the major obstacles your team faced in developing the algorithm?
So, when you're doing the initial drafting of a certain algorithmic approach, you want to see what each individual step does. And our visualization process was quite slow before…So that's something we just overhauled this summer, and hopefully that will mean that we can kind of speed up our algorithmic development.
There's also just so much other infrastructure work that needs to be done-- like data collection data processing, visualization work. And then aggregating census data into election data because they have distinct geometries that don't line up with each other.
What are you mainly focused on right now?
So right now, we're focused on that second approach I mentioned before, where we are generating tens of thousands of random districtings according to a non-optimal algorithm. We've really improved our optimal algorithm to be probably the fastest optimal redistricting algorithm that's out there. But we still have a lot of work to do with the random generation. So that involves a lot of other steps such as managing processes on the Stony Brook supercomputer, the Seawulf. To generate truly random samples would take a lot of computational time that our local machines just can't handle.
And that involves a lot of infrastructure work. That's something that was really the focus of our work for the past six months. So now it's about finding a version of the spanning tree algorithm that works best for throwing onto the supercomputer and creating a library of 10,000 redistrictings for every state that we have data for …and then after that, it's really up to what the data show to determine what we’ll be talking about in the future papers we hope to write . I'm really hoping that we'll be able to start publishing something once we get these libraries computed.
Yes, I'm really looking forward to the next six months of working on the projects.
What would you pinpoint as being most exciting about the work that you're doing currently?
I think the most exciting thing is that we're genuinely doing cutting edge work here. Obviously there are other groups and organizations and departments that are working on these problems, but it feels good to know that we’re one of the ones that are kind of pushing the fronts here on automated redistricting.
And, this is a very important problem right now… because in a couple months, the Census data will be available to everyone and the redistricting commissions will start doing the next round of redistricting in 2021, 2022. So it's really important that we get as much work out as we can, within the next year. Knowing that every minute I put into this matters right now because theoretically, that's a little bit more information for the redistricting commissions to use.
So that really helps me realize how important this work is. And it feels great to know that the work I'm doing is genuinely very important to democracy itself.
Have you done much research specific to New York or to Long island?
Not that much, because surprisingly, New York State has some of the worst data availability for elections. We've had an easier time getting data from the Georgia Secretary of State, for example. For New York State, you have to go to individually every individual county to get precinct level returns and updated geometry. So it's quite the process!
What are your long term plans for after you graduate?
I’m looking to apply for PhD programs next year. I’d be interested to keep working on the sort of problem I’ve been working on, discrete optimization problems, but something else I'm also somewhat interested in is natural language processing.
How was your experience of being a Teaching Assistant?
I've TA’d for a Theory of Computation class with Professor Bender. That was a great experience. One of the main benefits in the Honors classes is that there are ~ 30 to 40 students per class, which really means you get to know your peers and the students that are taking the class at the time. Everyone really gets a chance to kind of talk and ask questions. So it's a lot more interactive.
And being a TA has really helped me work on explaining things to people a little bit better. It's also shown me that if you want to be an effective teacher, you kind of need to understand how each individual student likes to learn, which is really instrumental for the work I'm doing on the PoliTech team. At this point, one of my responsibilities is making sure that the new students know what they're doing, and helping them to make contributions. There's so much infrastructure work involved with each aspect of the project that it really helps to have someone guide them, and help them digest it initially. So I think the experience of TA’ing has really helped me get better at that aspect of the project.
VIP seems to have been an amazing learning opportunity for you. How are you feeling about your overall Stony Brook experience so far?
I think that I think the Honors program in CS is one of the best things about Stony Brook. It's really just incredible; it doesn't just teach us how to be programmers and software developers, it really prepares us for tackling algorithm problems like the ones I've been working on at the PoliTech team. There's a large focus on the theory classes, which has been invaluable. I’m very happy here.