This post is a note-taking for the ‘Reading Research Papers’ part of the Stanford CS230 Deep Learning course lecture on YouTube URL. The class is all about advice on how to master a body of literature around some topics you want to learn. So I’ve written down his writings on the blackboard and his verbal comments.

  • Compiled list of papers(+ medium/blog posts)
  • Skip around list
    • (1) 10% of each paper or try to quickly skim and understand each of these papers.
    • (2) If based on that you decide that paper 2 is worthless, others say they sure got it wrong, or you read it, and it just doesn’t make sense. Then go ahead and forget it.
    • (3) As you skip around to different papers, it might decide that paper 3 is a seminal one and then it will continuosly go ahead and read and understand the whole thing.
    • (4) And based on that, you might then find the 6th paper from the citations and read that.
    • (5) Go back and flesh out your understanding on paper 4.
    • (6) And then find a paper 7 and go and read that all the way to the conclusion.
  • Number of papers you should read
    • 15-20 papers: good enough to do some work, and apply some algorithms.
    • 50-100 papers: enough to give you a very good understanding of an area.
  • Read 1 paper
    • The bad way to read the paper is to go from the first word until the last word.
    • Take multiple passes through the paper.
      1. Title/abstract/figures: the most
      2. Intro + Conclusions + Figures + Skim rest. (Skip/skim related work)
      3. Read but skip/skim math
      4. Whole thing, but skip parts that don’t make sense.
  • Questions for having good understanding of the paper
    • What did authors try to accomplish?
    • What were the key elements of the approach?
    • What can you use yourself?
    • What other references do you want to follow?
  • Sources of papers
    • Twitter (@kiankatan, @AndrewYNg)
    • ML subreddit
    • NIPS/ICML/ICLR
    • Friends
    • Arxiv sanity
    • (TMK) Papers with code (Web), Alpha zeta vector (YouTube)
  • To more deeply understand the paper,
    • Math
      • Redrive from scratch
    • Code
      • Download/Run open-source code
      • Reimplement from scratch
  • Longer term advice
    • Steady reading
    • Not shorts and burst

I’m grateful to Andrew Ng for sharing his wise and practical advice. As ending this lecture, he said, ‘some of this I wish I had known when I was a first-year Ph.D. student, but c’est la vie.’ I agree, but I think I’m lucky to know this now. It’s never too late to start. So keep going, and happy reading!