Ever wonder what grad school ai projects are like?
Here are my notes after taking a handful:
- pick a team of all students. no industry professionals (sorry, but let’s be
honest, your resources are constrained)
- gather the dataset and do something off-the-shelf
- start throwing random off-the-shelf algos
- start hacking at the data. split it up by feature, leave data out, etc.
- more hacky things. don’t worry about explanation or reproducibility
- even more hacky things. just do better than the other paper
- once you get a result, start writing the paper!! you have probably 2 days
left till project is due
- have a friend that has access to the packard basement printer. 2am printing
sessions are the best
Maybe it’s my perspective, but being a grad student isn’t very glorious. Implementation is hacky. Results are hard to reproduce. The problem is likely a toy problem, and if it’s not, real life datasets are very messy.
At best, your roc’s somewhere between random and slightly better than random. But you did better than random! You’ll get a resume project out of it and can say you contributed to research. Yesss, that’s exactly what I envisioned out of my grad school education.
Discussions
Discuss on Twitter