Based on S. Keshav's "How to read a paper".
Prevent drowing in details before getting a bird-eye view.
Summary
- Pass 1 - General Idea - abstract, headers, conclusions, figures, references
- Pass 2 - Understand - should be able to explain ideas
- Pass 3 - Virtually reproduce
- Also helps to learn how to present arguments
3 Passes
Pass 1
Expected time: 10-15 minutes
Actions:
- Read: abstract, introduction, section titles and subtitles, conclusion.
- Check references to see papers you've read.
Should be able to identify:
- Category - what does paper try to achieve? Propose new method? Summarize existing method? Measurements?
- Context - what papers does it reference? What main principles it relies on, methodology?
- Contribution - what are main contributions of the paper?
- Correctness - are assumptions that are made correct?
- Clarity - is paper well written
Enough for a paper you are interested in, but outside of your field.
Pass 2
Expected time: About 1 h.
Actions:
- Read carefully through the paper, but don't go deep into details of the proofs. Careful attention to figures and tables.
- Mark unread references for further reading.
Should be able to:
- Recount the reasoning of the paper to another person with supporting reference. You should grasp the concept of the paper.
Enough for a paper in your field, but not something you are directly working on.
Pass 3
Expected time: 4-5h for beginners, 1h for experienced.
To fully comprehend a paper you need a 3rd pass. It involved virtually re-implementing the paper.
Making same assumptions as the authors, re-implement the work.
By comparing this re-creation with the paper findings, you'll be able to identify the innovation, hidden assumptions and pitfalls.
Should identify and challenge every assumption in every question.
Also think how you yourself would present a particular idea. Comparing virtual and read leads to sharp insight in proof and presentation techniques implemented in the paper, you could add them to your set of tools.
Jot down ideas for future work.
After this pass you should be able to:
- Recall entire structure of the paper, its strong and weak points.
- All issues identified too. Implicit assumptions, missing citations, faulty experiments.
Kind of similar to Feynman method. Explain to yourself, come back, explain again.
Using for Literature survey
- Academic search engine with wll chosen keywords. Select set of recent papers. Do a first pass on them.
- Gives you a thumbnail of recent research in the field, and luckily a pointer to a survey paper.
- Otherwise, find shared citations and authors - leaders in the field.
- Find their websies and see what conferences they present at - those would be key conferences (as lead researches publish there)
- Find recent proceedings of the conferences. Select some of these. Let these and first pass papers consitute first stage of the survey. Do second pass on them.
- Identify key papers you've missed - locate and review.