🍀 How to Read a Paper

• Resources: [link]

🌿 Research papers are the first rough draft of science 🌿

🍀 Motivation: Why read papers

• To learn about ideas in an area
• To learn about the community of people in your area
• To understand the trajectory of research
• To write papers
• To debate papers in a certain course :)

🍀  Coarse-to-Fine Reading: Read the paper in multiple passes

🍀  Coarse Reading

Blackboxing

• Print the paper. Make notes on the paper.
• Read the paper quickly the first time (in an hour)
  • Pick up the notation
  • Understand the inputs and outputs (blackboxes) of the paper, the sections, etc.
  • Note all the claims of novelty in the paper
• If I don’t know something, I quickly look it up; if I still don’t get it, I move on.
• Most authors write as if the readers already understand the paper as well as they do. This heuristic gives you an approximation of that understanding.
• In well-written papers, the figures and captions tell you how to blackbox the paper. Well-produced SIGGRAPH movies help with blackboxing.

🍀  Fine Reading

Can I code the idea?

Think like you’re coding the idea! Understand Everything in the Paper → Everything

• Read the paper a second time much more carefully, working out what each equation means
• Do not move until you get each equation well enough to code it. If you’re stuck look it up. Discuss it with your lab mates. Discuss it with the faculty. Discuss it with me.
• Approach the whole paper as you would if you were coding it (sometimes I actually do prototype some idea kernels)
• Keep re-reading until you get everything.

🍀  Finer Reading

How did the authors come up with the solution?

• Be critical → Run a simulation in your head of each idea
  • Is it original?
  • Is it reproducible?
  • Did the authors omit any results? Any figures?
  • Were all the claims justified?
  • Is it likely to inspire future work?
  • Are there simpler solutions to the same problem?
  • Did they evaluate the approach thoroughly?
• Be constructive → Run a simulation in your head of each idea
  • When you get proof, accept it
  • (Almost) all ideas are obvious in retrospective. Was it trivial to come up with the idea prospectively?
  • What are the contributions of the paper?
  • How much of step forward is it from previous work?
• Multi-pass → Keep reading the paper until you get it
  • The trajectory of ideas: read the citations you marked in the first two passes
  • Make connections to internal and external ideas
  • Create a narrative for the paper
  • Think about the solution until it becomes obvious why they chose to do it the way they did

🍀 Finest Reading

Why did the authors come up with the question?

• Read widely and deeply; Think widely and deeply; “Do” widely and deeply

🍀 Miscellaneous Advice

• Print the paper. Makes notes on the paper.
• Be very conscious about what you like/dislike about the paper
• Live with the paper for a few days. (Greg Turk)
• Mark interesting citations for future reading.
• Discuss papers with lab mates. Discussion is a form of thinking.
• Find the easiest access point. Link it to ideas you know about. (Simon Lucey)
• Don’t be intimidated by math. (Iain Matthews via Simon Lucey)
• Don’t be biased prematurely by the results. (Jason Saragih)
• Read a lot. Build your paper classifier. (Kris Kitani)

🌿 Be patient with yourself. 🌿

🌿 Reading papers takes time. 🌿

🌿 Read, Read, Read: build a foundation of ideas. 🌿

🌿 Develop rapid-reading skills: connecting to ideas you already have. 🌿