(Last updated: June 10, 2021)

Over the course of this summer, I intend to read many papers related to quantum computing and federated learning. So far, most of my blog posts have been about articles and tutorials published online. However in the next few days, I hope to share a few posts about peer-reviewed papers. These are generally more complicated to read through and understand, and they are hard to write about. That’s why I’d like to share “How to Read a Paper” by Srinivasan Keshav, now faculty member at the University of Cambridge. This paper was published by ACM SIGCOMM Computer Communications Review in July 2007.1 It provides an algorithm for reading papers efficiently and effectively.


Introduction

Reading papers is an important skill for researchers that is rarely taught. Instead, graduate students jump into the fray, wasting much time learning from trial and error. This paper attempts to reduce the learning curve of how to read papers in the most efficient way possible.

Three-Pass Algorithm

Given a paper, you should plan to read it three times:

First Pass (5-10 minutes)

Without reading details or illustrations, get a general understanding of the paper’s topic and how it will approach the topic. Determine the paper’s category, context, correctness, contributions, and clarity (the 5 C’s). After this pass, you should be able to determine whether or not the paper is worth the second pass.

  1. Read title, abstract, introduction
  2. Read section, subsection headings
  3. Read conclusions
  4. Glance at references

Second Pass (less than an hour)

Scan through the paper to get the general gist. Afterwards, you should be able to write a brief summary on the contents. Then you can decide whether or not you should continue to the third pass.

  1. Read paper with pencil (ignore proofs)
  2. Carefully go through figures, digrams, and illustrations
  3. Mark any relevant references you haven’t read yet

Third Pass (inexperienced: 4-5 hours, experienced: 1 hour)

Carefully read through the paper and attempt to re-create its content and conclusions. At this point you will be able to identify strong/weak aspects, assumptions, issues, etc. It should also be partially committed to memory.

  1. Virtually re-implement the paper

Completing a Literature Survey

The key for completing a literature survey is to find a collection of highly-relevant and recent papers contributing to the field and to perform the Three Pass algorithm smartly.

  1. Search for keywords in an academic search engine (e.g. Google Scholar, CiteSeer)
  2. Find 3-5 recent papers and perform the First Pass on each of these. Look for a pointer to a recent survey paper (note if you find this survey paper, you can skip the following steps).
  3. Identify shared citations to determine key papers/researchers in the area
  • Download papers for future reference
  • View websites of key researchers to identify top conferences in the field
  • Visit websites of top conferences and look for recent work in field

References


  1. Keshav, S. “How to Read a Paper.” ACM SIGCO>MM Computer Communication Review, vol. 37, no. 3, 2007, pp. 83–84., doi:10.1145/1273445.1273458. ↩︎