My advisor used to tell me: read broadly to gain knowledge; read with questions in mind; grasp the main message from papers.

While I totallly agree these sounds pretty good, I don’t really know how to “implement”.

It usually took me an hour to read a paper from front to end while taking lots of notes. Then a week later, I remember nothing about the paper and seldom check my notes.

I don’t feel productive while reading like this.

I thought the reasons my reading is slow and not effective are:

  • my English reading is not fast enough

  • my bad memory

I thought the only way to go is to read more, take more notes and review notes. So I did, but that did not solve my problems.

Thankfully, I came across these 2 articles which changed the way I think about reading:

https://blog.usejournal.com/advice-on-building-a-machine-learning-career-and-reading-research-papers-by-prof-andrew-ng-f90ac99a0182

http://www.pgbovine.net/opportunistic-paper-reading.htm

My takeaway is that: the only way to read a lot and remember the key message of each paper is to read in a “bread-first” approach.

I have started doing this and it made my life so much easier.

  • Do not care about the techinical details and do not take notes while reading the first time, after reading, close the paper and think about “what did I learn that is relevant to my project?”, then take note of that.

  • Read 20 papers like this on a research topic can be done easily and give a reasonably sense of what questions people care about, what approaches people use on this topic.

There’s also a issue of reading too much. As Phillip Guo said in the article linked above

The more papers you read, the more discouraged you will grow because you'll come to realize that every single idea you're thinking about has been done before in some other form. No idea is truly original. Senior researchers know how to overcome this paralysis, but for someone who is just starting out in research, this discovery can be crippling. It might even make you want to quit and switch research areas ... "

I realized I have this problem since I started following researchers I admire on Twitter. Their insights and good creative taste always impress me but also seeing their tweets make me feel “oh they are interested in this problem as we do… Oh they published this model! … oh that’s so elegant, I can never do that… oh I should quit this idea…”

Now I decide to start follow Phillip Guo’s advice: First start building up some momentum on your research project for at least a few weeks…just first get started on a somewhat promising path. After you’ve built up momentum, start reading.

I learned the lesson: do not start searching and reading when first starting on a problem! Because in my case, that’s very likely to do more harm than good for my research progress.

I wish I knew this earlier in grad school…