Each year, OpenDP offers a summer internship to undergraduate, post-bacc, or graduate students for a 10-week program working on various projects under the OpenDP initiative or the broader privacy tools project.
We have eight excellent students joining us this year that we wanted to highlight in a series of blogs to get to know them better. First up, we have Patrick Song and Vicki Xu!
Patrick Song
Q: Tell us a little about yourself
I'm a second-semester senior at Harvard studying computer science. On campus, I help lead the meditation club and enjoy playing cello in orchestra and various chamber music groups. I also enjoy running, hiking, (I just spent a wonderful week exploring Jasper and Banff National Park in Canada!), and watching my favorite Boston sports teams.
Q: What are you currently working on?
I'm currently working on a Human-Computer Interaction (HCI) study investigating user mental models of open-source DP libraries. As part of the study, I'll be interviewing library developers from different organizations (OpenDP, IBM, Google, Tumult Labs) and will also conduct user studies comparing how users (without significant experience with DP) interact with different open-source libraries, exploring how privacy concepts are reinforced through design and implementation choices that library developers make. I'm also interning on the data and privacy engineering team at Patreon this summer, and am excited to learn more about data privacy practices in industry.
Q: How does that support OpenDP?
Although DP has gained prominence in industry and academia over the past several years, there has been little evaluation of the extent to which open-source DP libraries are helping engineers and scientists achieve their data privacy goals. Given the challenges in translating theoretical DP concepts into implementations of DP algorithms, there is a need to examine the utility of open-source libraries for data practitioners. Results from this study will hopefully offer insights into best practices for developing intuitive DP applications, which will support future development for OpenDP and other open-source privacy tools.
Q: What are you planning to do after graduation?
I'm currently in the process of applying to law school (exploring the intersection between tech/policy/law is of interest to me), which I hope to attend after spending a couple years as a privacy engineer in the tech industry.
Vicki Xu
Q: Tell us a little about yourself
I’m a second-semester senior at Harvard College studying computer science and math. Salil is my thesis advisor. I’ve been working with OpenDP since last summer — previously, I was working on the proofs of the theoretical accuracy of the primitives in the OpenDP library with Hanwen Zhang. In my free time I like to hike, cook, paint, and read and write fiction.
Q: What are you currently working on?
I’m working on concurrent privacy-loss odometers and filters, which is what my senior thesis will be on. This project will have both a theoretical and a practical element.
Q: How is that impactful for OpenDP?
Privacy-loss odometers and filters basically give a data analyst a way of gauging how much an algorithm (or set of algorithms) on a dataset might erode the privacy of the members of that dataset, over the course of computation. You can think of a privacy-loss odometer in the same way you think of the odometer on your car, the mileage being privacy loss (which is a quantity that depends on which variant of differential privacy you’re ascribing to, e.g. epsilon-delta, Renyi, etc.) — i.e. it’s an add-on to an algorithm that gives you the privacy loss at that time if you ask for it. A filter is also an algorithm add-on, but it forces the algorithm to halt if privacy loss exceeds a certain budget you give it. Prior theoretical work in this area has primarily focused on odometers and filters of sequentially-composed algorithms, in which the analyst interacts with a set of algorithms in sequence, but I’m studying the concurrent setting, which is when the analyst can essentially revisit algorithms it has already queried. This is helpful for users of the OpenDP library because it offers them a way to gain more visibility of the privacy loss over the course of computation.
Q: What are you planning to do after graduation?
What a great question! I’m applying to some fellowships right now. Since I have my spring semester off, I’m also planning to travel during that time. But there are a couple of large decisions I need to make this summer (in particular, whether or not to start work right after college) and I’m welcoming advice on how to go about making them :)
