How to read papers?

Submitted by chaky on Wed, 11/02/2022 - 13:09

Reading every sentence is time-consuming.   Here are some tips:

  1. Mindful of the publication date:  This will slowly help build you a good history of papers in your mind.  (but slowly...)
  2. Look at figures first:   Get a feeling by scanning the figures.  Good papers usually have good figures, which would give you a good feeling what this paper is about.
  3. Skim the whole paper:   Do not read words by words.   Go to the methodology first.   Methodology is one of the most important part to read because you as a researcher can learn and copy.  In the beginning, you don't have to read every sentence, just identify what is the key things in the methodology.   Then quickly jump to results and discussion to check the key findings.    Then you can go back to introduction to understand the context of this research.   
  4. Remember the authors:   It's nice to remember great authors to expand the pool of experts you know.
  5. Look at the references:   Look at the references, check whether there is any paper interested to you.  Do 2-3-4 again.
  6. Make a summary table:  After skimming like 25-30 papers,  create a table summarizing each work, and try to identify the gap or limitations.   Identify a list of potential topics you can work on.    You can also start writing your Related Work section of your paper, to become more organized.
  7. Identify "key" papers.  You found these 3 papers to be something close to what you want to do, read them over and over again, until you understand by heart.
  8. Code / Experiment.   After you understand them by heart, it's time to try to emulate them by coding or doing some study.   This will make you even understand better.   Sometimes the author provides a GitHub so you can copy and learn.
  9. Test your ideas.   After you are able to emulate some work, you may want to change to your own new idea.  See if it works.  Test many ideas.    Finally, once you found some pattern, you can tell your professor that you found something interesting...