Research philosophy

In this post, I pompously share my thoughts on how I aspire to be a researcher. My hope is that junior researchers who hold similar ideas and have faced the challenges that come with holding them, feel some sense of support from reading this. I’m always happy to chat more about navigating these challenges amidst the rat race that is AI/ML research.

  1. I seek simple, insightful results unburdened by mathematical obfuscation. This means pursuing minimal abstractions, example and counterexamples, vivid intuition, clarity and depth of thought and nuanced argumentation.

  2. Researchers should be publishing significantly less. In the first ten years of my career (2015 to 2025, the date I write this) I have published about a paper a year1, I wish it was less frequent.

  3. Peer review is a form of taxation. If you publish more, you should review more, while ensuring that your reviews are of high quality. The System Is Broken™ in part because everyone’s publishing more without paying their fair share of taxes (and also because there’s more of everyone).

  4. I care about clear and accessible communication. I enjoy spending an unhealthy amount of time on making my talks and writing my papers. This is a constant learning process—every paper or talk takes at least two or three attempts to attain clear presentation. There is significant insight to be derived from this process.

  5. I enjoy the detective work that goes behind cataloging an extensive, well-organized related work section. It’s a satisfying process. If there are missing citations in my work, please reach out to me so I can rectify it. Again, there is significant insight (and knowledge) to be derived from this process!

  6. I believe that science must be done and truth must be sought collaboratively with compassion and good faith, building on each other’s works and strengths.

  7. Science is more subjective than it is made out to be. We must embrace this with humility. It’s not always obvious what questions are important, or what directions will be fruitful.

  8. At this point in my field, the only form of external validation that matters to me is when others tangibly build on my work, or if someone enjoyed reading my work. All other metrics (including citations, publications, awards, talk invitations and fellowships I’ve listed elsewhere on this site) have become utterly meaningless.2

  9. Byte-sized dissemination of research on twitter has been a terrible idea for science (but I do it anyway). It has encouraged reductive and combative interactions. 2

My values are shaped by what I’ve read.

Footnotes

  1. As lead contributor or mentor. As I grew in seniority, I’ve additionally participated as a “middle author” in about a paper a year. ↩

  2. Yes, that’s hypocritical! But also see the Tu quoque fallacy. ↩ ↩2

« Why PhD students should read the history of science Angles between high-dimensional vectors »