Reference: https://codearsonist.com/reading-for-programmers

3 pass Overview:

  1. Verify
    1. Value
    2. Background
  2. Grasp
    1. Thought Process
    2. Highlights
    3. Diagram
    4. Unknowns
    5. References
  3. Understand
    1. Holes
    2. Implementations
    3. Notes

Process

Verify

  • This provides High level view of the contents.
  • Start with references usually.
  • Mark the cited papers I’ve read as to get previous notes I have made and
  • verify the background.
  • Title, Abstract, Introduction–> Read very carefully
  • Scan through to find diagrams, equations.
  • Questions to ask during first pass

    • What type of paper is this?
    • Context: How is it related to and which other papers?
    • Correctness: Do assumptions appear to be valid?
    • Contributions of the paper
    • Clarity: well written?
  • Now you ask yourself?

    • Do I have enough background?
    • What’s in it for me ?

    [Note: If I don’t have enough background, put a TODO with :advanced: in it.]

Grasp Concepts

  • To understand and explain to others as well it’s evidence and concepts
  • Read paper , scribble through, make highlights.
  • Inline references : some are instant hook: added to TODO list.

Critique

  • For highly interesting papers I need to comprehend
  • It’s said you need to recreate a paper without copying in order to review a paper.
  • For papers vital to work–> Pick tools you use and recreate it.
  • Bits of code and notes.
  • Critically look for inconsistencie

Note Taking Process

  1. Compile list of papers, medium articles and resources.
  2. Skip around the list in parallel. Skim and understand 10-20%. By this time, decide which ones are worth spending time on.
    1. For knowledge of field and implementation: 5-20% comprehension required
    2. For good understanding of domain application: 50-100%
  3. Read paper, skip math initially.
  4. If familiar with the field, go through Related Works section directly.
  5. For maths:
    1. Try to rederive from scratch.
    2. Find Open source codes if possible
  6. To get depth in it:
    1. Make open source contribution, do some related project or any research and internships.

References

(Daniel Adiwardana et al., 2020)