source

How to Read a Paper

What is the Problem?

Researchers spend 100s of hours reading papers every year, so they ought to know how to do it effectively. Keshav argues that reading a paper is a skill that can be learned, and that it is a skill that is not taught in school.

Summary

Keshav recommends a three-pass approach to reading a paper:

First Pass

Gives you a general idea of what the paper is about. Read the title, abstract, introduction, and conclusion. Only read section and sub-section headers. Also glance over the references and note which you've read. This should only take 5-10 minutes.

After the first pass, you should be able to answer the following questions:

• Category: What type of paper is this? (e.g. measurement, analysis)
• Context: Which other papers is it related to? What is the theoretical background?
• Correctness: Do the assumptions appear to be valid?
• Contributions: What are the paper's main contributions?
• Clarity: Is the paper well written?

Second Pass

Grasp the content of the paper, but not necessarily the details. Read the whole thing, but ignore things like proofs. It helps to also annotate/take notes during this pass.

Pay special attention to diagrams, and also mark relevant unread references. This should take about an hour.

After this pass, you should be able to summarize the paper in a few sentences to a peer. If you still don't understand, you can do one of three things:

• Set the paper aside and hope
• Come back to the paper later after looking up references you didn't understand
• Continue to the third pass anyways 😢

Third Pass

Understand the paper in depth. You should virtually re-implement the paper, following all reasoning and challenging every assumption. Also try to think about how you would have presented the information differently. You should note potential follow-up work during this pass. This should take up to 5 hours, and at least 1 hour for a well-written paper and well-read reader.

After this pass, you should be able to reconstruct the paper's structure from memory. You should also be able to critique the paper and pinpoint implicit assumptions, limitations, and potential improvements.