How I Read a Paper Without Falling Asleep

Early in my PhD I tried to read papers the way I read books. Start at the top, finish at the bottom, understand everything. It took hours per paper, I retained very little, and I felt vaguely guilty all the time.

The problem was treating every paper as if it had earned my full attention. Most have not. That is not arrogance, it is arithmetic. Thousands of relevant papers exist and your afternoon does not scale.

Pass one: five minutes, decide if it is worth more

Title, abstract, figures, conclusion. That is it. You are not trying to understand the paper. You are answering one question: does this matter to what I am doing?

Read the figures before the text. A good paper's figures carry the argument, and a paper whose figures make no sense on their own is often a paper whose argument does not either. Most papers stop here, and that is a successful outcome, not a failure.

Pass two: thirty minutes, get the idea

If it survived, go for the actual contribution. What did they do, what did they compare it with, and what is the evidence?

Skip the related work section on the first read. It is usually written for reviewers, not for you. Go to the method, then straight to the results, then check whether the results support what the abstract claimed. That last check catches more than you would like.

Questions I keep in the back of my mind the whole time:

• What is the baseline, and is it a fair one or a deliberately weak one?
• How big is the dataset, and how was it split?
• Would this result survive on data from a different hospital or a different device?
• What did they choose not to show me?

Pass three: a few hours, only if you need to build on it

Now you go line by line. Reconstruct the method as if you had to reimplement it, because you might. Check the equations. Look for the assumption they made quietly in one clause and then never mentioned again.

This pass is expensive. Very few papers deserve it. The ones that do are usually the papers you are directly extending, or directly disagreeing with.

Write something down

Two or three sentences, in your own words, right after you read it. What they did, what surprised you, whether you believe it. If you cannot summarise it, you did not understand it, and it is better to discover that now than in a viva.

This is also, in case you were wondering, the entire reason I ended up building a tool to keep track of all this.