Austin Watkins
Malone Hall 3400 N Charles St
Baltimore, MD 21218
Hello, I’m Austin and I’m on the job market! I’m finishing my PhD in computer science in 2027 at Johns Hopkins University, advised by Prof. Raman Arora. I do research in machine learning theory: multi-task representation learning, differential privacy, adversarial robustness. I’m particularly proud of some of my recent work. My first paper at NeurIPS shows it is possible to leverage multiple tasks to provably few-shot learn by learning an embedding model, and my second at NeurIPS shows that it is possible to do so while protecting against adversaries. My latest preprint shows that via perturbing the objective it is possible for gradient descent to be differentially private with only one draw of Gaussian noise instead of one per step as has been the case since 2016. Lately I’ve been doing empirical research in AI security: in 2026 I designed a novel adversarial attack on language models and disclosed it to the affected AI labs.
My CV is here.
I completed an M.S.E. in computer science at Hopkins, a B.S. in applied mathematics, and a B.S. in computer science at the University of Utah, the last being advised by Prof. Jeff Phillips.
Not only do I have a foundational understanding of machine learning but I also can ship code. I have an unusually large amount of software engineering experience for a “theorist.” I was a software engineer intern at Amazon Web Services where I implemented an interface to communicate performance and health analytics of the DynamoDB fleet such that it can be visualized and mined. I received a full-time offer, perhaps because I was asked to implement a few endpoints but instead built a general layer covering nearly all of them, that or my mad Vim skills. Yet most significantly, in 2019 I cofounded a company and functioned as its CTO. I took what was originally a 1,918-line prototype god function with unreachable control flow and turned it into a tested, linted, and containerized codebase. I managed the databases and wrote the service layer, nearly all the REST API, and the forecasting pipeline. Besides learning a lot about building a culture that cares for the code and shipping features under pressure of a client I enjoyed it tremendously.
I had an atypical education as a child. I don’t have a high school diploma or a GED. Nor have I taken the SAT, ACT, or GRE. As a kid, I watched a lot of Arthur and played many video games. I also spent a significant amount of time on my hobbies of card magic, chess, being on a competitive dance team, and reading prepper and survival books. When starting adulthood, given the situation, I did a lot of menial labor. In 2015 I was selling solar panels door-to-door in the Las Vegas summer. I wanted better for myself. That year I decided to close the gap. On Khan Academy I watched every video and did every exercise to mastery from kindergarten to 12th-grade mathematics in a little over a year. Later in 2015 I read Tim Urban’s Wait But Why series on AI. There, I was influenced by Müller and Bostrom’s survey, where experts expected “high-level machine intelligence” within my lifetime. I concluded that if transformative AI came so soon, I wanted to play a small part in it. The next day I Googled, “who does research in AI,” then committed to a PhD in ML. Eleven years after reading that series I have published at NeurIPS, have nearly attained my PhD, and still aspire to play a beneficial part at the frontier of AI.
I have read Ryan Holiday’s The Obstacle Is the Way many times. For a while, every time something “bad” happened, everything from heartbreak to career setbacks, I read the book. The chapter titled Iterate begins with the quote by Wendell Phillips, “What is defeat? Nothing but education; nothing but the first steps to something better.” I believe my life has embodied those words. My skills and creativity will be an asset for my next adventure, whatever it may be, as seen in the micro of designing a novel adversarial attack on language models and inventing a new way to ensure privacy for people and in the macro of going from studying elementary math in 2015 to being nationally recognized for my undergraduate research five years later to soon finishing my PhD.
I also enjoy long walks, cooking, photography, reading, magic tricks, and performing improv (indie team; trained at the Baltimore Improv Group) and stand-up comedy in Baltimore.
news
| October 2026 | 💼 I’m on the job market! I expect to finish my PhD in 2027. |
|---|---|
| October 2026 | 📖 I completed the BlueDot Technical AI Safety course. |
| October 2026 | ⭐️ New preprint: Differential Privacy of Gradient Descent on Perturbed Objectives, now on arXiv. |
| May 2026 | 🏆 I was named a Gold Reviewer for ICML 2026. |
| October 2025 | 🏆 I was named a Top Reviewer for NeurIPS 2025. |
| January 2025 | 📚 I’m the head teaching assistant for the Machine Learning class taught by Raman Arora. |
| December 2024 | 🎓 I completed an M.S.E. in computer science at Johns Hopkins. |
| September 2024 | ⭐️ Our paper, Adversarially Robust Multi-task Representation Learning, has been accepted at NeurIPS 2024! |
| August 2024 | 📚 I’m the head teaching assistant for the Learning Theory class taught by Raman Arora. |
| May 2024 | 💻 I’m doing an internship at Booz Allen Hamilton. I am a scientist intern doing internal research on differential privacy. |
| October 2023 | 🎤 I presented my work on adversarially robust learning at the DARPA GARD program meeting at IBM. |
| September 2023 | ⭐️ Our paper, Optimistic Rates for Multi-Task Representation Learning, has been accepted at NeurIPS 2023 as a poster! |
| August 2023 | 📚 I’m a course assistant for the Machine Learning class taught by Raman Arora. |
| January 2023 | 📚 I’m a teaching assistant for the Machine Learning class taught by Raman Arora. |
| August 2022 | 📚 I’m a course assistant for the Machine Learning class taught by Mark Dredze and Anqi Liu . |
| June 2022 | ⭐️ My first publication! Using Existential Theory of the Reals to Bound VC Dimension has been accepted to the Canadian Conference on Computational Geometry. |
| January 2022 | 📚 I’m the head teaching assistant for the Software Testing and Debugging class taught by Ali Darvish. |
| September 2021 | 🏫 I’m now at Johns Hopkins University as a computer science PhD student researching machine learning! |
| January 2021 | 🔬 I’m a predoctoral fellow at the University of Utah hosted by Jeff Phillips working on adversarial robustness. |
| December 2020 | 🎓 I have graduated with honors with a B.S. in computer science and applied mathematics. My thesis, in the computer science department, is on using tools from algebraic geometry to bound the VC dimension of inflated polynomials. |
| December 2020 | 🏆 I have been recognized by the Computing Research Association with a 2021 Outstanding Undergraduate Researcher Award - Honorable Mention. |
| May 2020 | 💻 I’m doing a summer internship at Amazon Web Services. I am on a Python-based data science team within DynamoDB. |
| December 2019 | 🚀 I co-founded Kytheran Labs, where I’m the acting CTO. |
| August 2019 | 📚 I’m a teaching assistant for an intro to theoretical computer science class taught by Ganesh Gopalakrishnan. |
| January 2019 | 🔬 I have started research on the numerical error in neural networks with Prof. Ganesh Gopalakrishnan while being mentored by PhD student Vinu Joseph. |
| January 2019 | 🔬 I’m now doing undergraduate research in financial machine learning with Jingyi Zhu in the Math Department. Among other things, I wrote a technical report with Carlos E. Jimenez. |
| January 2018 | 🏫 I’m now attending the University of Utah! |
| December 2017 | 🎓 I have graduated with an A.S. in General Studies with a 4.0 GPA while working 29 hours per week at the college graduation office. |
| January 2017 | 🏆 I have been awarded the Pride In Academics award. “Awarded to accomplished students of diversity,” due to my academic accomplishment while being a first-generation student without a high school diploma. |
| May 2016 | 🏫 I’m now attending Salt Lake Community College! |
| December 2015 | I have left my sales job. |
| October 2015 | 🤔 I have made the goal to do a PhD in machine learning. |
publications
-
NeurIPSAdversarially Robust Multi-Task Representation LearningConference on Neural Information Processing Systems, 2024
-
NeurIPSOptimistic Rates for Multi-Task Representation LearningConference on Neural Information Processing Systems, 2023
-
CCCGUsing Existential Theory of the Reals to Bound VC DimensionCanadian Conference on Computational Geometry, 2022