Mayank Agrawal

mayank dot agrawal at princeton dot edu

what i worked on

(modeled after Paul Graham's What I Worked On)

In high school, I could sum up my 'extracurriculars' as teaching and running. But before that, middle school was probably when I had my first maker role. I made sweet (shitty?) Youtube mixes of NBA players. Yao Ming, Von Wafer, Kevin Durant, and I think a fourth. I found a way to get NBA highlights and I would mash them up to music and post them on YouTube. I learned search engine optimization from my uncle, and I used that know-how to rank my videos well. I was obviously the only one that made a mixtape of Von Wafer, and so I was the first on results. The video got over 100K views because he was signed by a team in Greece and everyone started looking him up. I made videos a couple times afterwards, a couple in High School for Art and Spanish class, and one in college for Linguistics. I really enjoyed them. The Linguistics one was neat because I got to interview many of my friends about their experience growing up in bilingual environments.

Teaching was big for me, first as a way to `give back' and then I realized I was really good at it. It all started with Math and Science Team tutoring, and it then turned to Breakthrough. I became a core part of of Breakthrough because I loved those kids and they loved me. I was real with them, and they appreciated my authenticity. I became a damn good teacher, too. I learned how to expertly structure lessons and how to manage interpersonal classroom dynamics. Stuff went well TAing at Swat, and the 3 classes to TA at Princeton were good, too, sans the Zoom.

Running was obviously a big part of high school and college (and early graduate school). It's a story I've told often, but doesn't really get old. I started off as a scrappy kid with a chip on his shoulder. I ran a bit, got Sumant to hook me on running everyday. Eventually, I joined track, met up with Erik, he introduced me to Andy and his training group, and the rest is history. I made so much progress, had incredible teammates, and was just genuinely really happy with life. It gave me a sense of meaning and purpose, as well as a wonderful community. Running in college was great, too, sans the team drama that enveloped me by the end. But, runners and past teammates continue to be some of the closest friends I have ever had, and I still have a dynasty fantasy football league going on with people from Swat!

I worked at Mercury Fund the summer before I went abroad in college. I had also worked at Carnegie Mellon that summer, and UPenn the summer before, but both summers were horrible work-wise. I had no idea what to do and was just lonely and flailing around. In retrospect, maybe this was a sign that academic research environments weren't the best for me. Mercury Fund was fun, though - except for the part when I had to go that conference. I had good mentorship, made a fun taxonomy of the data science companies, and taught the partners the fundamentals of data science. It turns out, I indirectly helped the incubation of Mercury Data Science.

Then there was Kallyope, a biotech startup in New York City. I worked on the computational team under Justin, who everyone was afraid was but I liked (because he was intense). I pushed myself hard that summer (or at least the same level of difficulty I usually push myself) and apparently he was quite impressed with what I did. We created this whole pipeline to predict the proteins that would be created from sequencing data, and we did this all my trying to recapitulate the biological pipeline in terms of state-of-the-art neural networks. At the end of the summer, Justin said I probably had done enough work for 2 Nature papers, but who knows if he was just stroking my ego.

There a few other cool school projects that I distinctly remember. One was my 'Extended Essay' in high school, where I combined Moneyball / NBA Statistics and multidimensional geometry. It was probably bad, but it was really fun. Mr. Guidry had my back when I proposed this to the math department (who never has anyone try to do an extended essay). It got a low grade by the higher IB powers, and was probably botched execution, but it was fun. The general idea was to consider every member of a starting lineup as a point in N-dimensional space, where each dimension was a z-score value for a specific stat, and then compute the volume for this 'simplex', with the hypothesis that teams with a higher volume would win more because teams with a higher volume signified complementary skillsets. I'm pretty sure talking about this essay got me into Swat.

Another fun project was at Swat, where I worked with Owen and Oscar to make GreenMon. It was a project we did because we thought it would be an easy grade (it was), but also because it was fun. The idea was to build some layer over this peer-to-peer monitoring service to track cluster usage, and dynamically start and stop nodes in order to save energy. It actually ended up being useful, and we ran some tests that said it was would work really well. It was never meant to actually be deployed, but who knows. It was a fun team, too.

This now reminds me of my college thesis, which again was botched execution probably, but was fun and I think got me the PhD offer from Jon. The idea was initially inspired to respond to Sharon Street's evolutionary debunking argument against moral epistemology - essentially, that we should not think we can learn moral truths because there is no reason to think our learning process selected for that. I was learning computational learning theory at Oxford, which is machine learning meets computational complexity theory, and so my thesis was to try to formalize moral learning as a reinforcement learning problem and talk about the different complexity demands of this process. It was fun. Very handwavey and definitely pseudo-intellectual in retrospect, but it was fun and I think it impressed my Swarthmore advisors.

And finally, Princeton. First, there was the new theory for cognitive fatigue, which I still remember the 'Aha' moment for. The idea was to map cognitive fatigue onto hippocampal replay, and we got lucky that down the hall, Nathaniel and Marcelo had just dropped an incredible theory of hippocampal replay. We all joined forces and then published this amazing paper in Psychological Review. The fMRI project afterwards has been a mess (coinciding with COVID and whatnot), but what a wonderful theory paper.

The work with Josh and Tom also turned out really well. Tom first told me to look into the Moral Machine dataset because of my supposed interest in morality (long story). We transformed it into a nice methodological paper, which outlined how to do science in a really novel way that leverages big data and machine learning. It was published in PNAS, and we followed up with the Science paper, which combined the original work they were doing with risky choice with the mixture of theories we found with Scientific Regret Minimization. A seminal work for sure.

Lastly, there was the philosophy paper I worked with David Danks trying to combine consequentialism and deontology using bounded rationality. The idea here was relatively simple but powerful: consequentialism is a computationally intractable problem and people should use rules (heuristics) to approximate rightness when calculating is not possible. This gives interesting predictions that people should be consequentialist in high-stakes settings and deontological in low-stakes. I think this paper was the teaser that potentially got me the faculty offer because it really showed novelty and originality.