I didn't see if this question has already been discussed, but what are your techniques/strategies to read papers? Is reading a paper to find the next problem you'll address different from reading a paper to write a review? Do you write a summary of each paper you read to remember it later (if so, what kind of summary? Slides? A short text?)? How much time do you spend on reading a paper? Do you do multiple passes?

Lots of questions! I will say that reading a paper is a skill that takes time to develop. My first year in grad school, it might take me 4-5 hours to read a paper and I'd have to take detailed notes, etc. After 20 years of reading papers every day (more or less), it's a lot easier.

Definitely read in passes, depending on what you're trying to get out of it. When I review a paper as a member of a PC, it usually takes me about 90 minutes to read the paper and about half an hour to write up my review. If I'm reading a paper just to get a sense for if it is relevant to read further, it might only take 1-2 minutes

I've never found notessummariesetc useful, personally. In part because what is relevant about a paper changes over time. Maybe the first time I read the paper I'm interested in generally learning the concept. Later I might really be interested in understanding how they modified part of the OS to do something. Later I might want to read the paper because I need some examples of good workloads for evaluation. So much like how people can get different things from the same poempaintingbook/etc, over time.

As with everything, though, YMMV.

For me personally one method of reading papers that helps is to do multiple passes. I typically do it like this: Read Abstract, intro and conclusion highlighting the claims the authors make. Then I spend perhaps 10 minutes just leafing through the rest of the paper. I let it “rest” for a day I refresh my memory with the highlighted aspects and then I read related work and background and the implementation details. I highlight important points and mark questions on the side of the sheets. I let the paper rest for a few hours. I read evaluation and discussion, implementation carefully and try to find answers to my pending questions. After that I recycle the sheets, I rarely keep notes around. I find that reading the paper like this helps remove any subjective bias (or anger :slightlysmilingface: ) from the process and I can usually learn or understand more than If I try to go at it in a single sitting.

There is a relatively famous write-up that suggests to use passes (link). It works for some people and does not work for others. I don't think that is the “best” or the only way to do read papers and that one does not work for me. As it is quite hard for me to read only part of the paper and abandon it as well as do multiple passes I found that a different strategy works better for me. I prefer to read the paper once from start to end. I used to write a very brief (couple of sentences) note about the paper as I will forget almost everything about the paper but that didn't work so I started to use longer notes that I can keep and re-read once needed (think brief summary with main points). I like the Cornell note-taking system so I created a template recently based on it. I have there section for taking notes during reading the paper and after I'm done with the paper I will fill the part about main ideasresultsstrengthsweaknessesetc. If you want to take a look below are the links to the Google doc version and MS Word doc version (works better with table alignmentssizes). Feel free to useshare. Google Doc version link MS Word docx file link If your goal is more about learning papers instead of reading them (I don't know how many people distinguish between those two) then you can also look into a method called “spaced repetitions” and the software that uses that method is called Anki (I believe there are others). It is used for flash cards but ultimately you can make your notes to be flashcards. (edited)

I use Anki for learning languages, but I didn't think of using it for reading papers :smile:

I tend to skip the abstract, intro, and motivation to see if the paper solves a problem I care about. If it does, then I will read more. Generally, I set a timer and force myself to take no more than 20 minutes. At the end of that, I usually have a reasonable idea of what's going on. Generally, I don't read much of the implementation, evaluation details, etc unless I have a specific question/concern from reading the design. Then, I write 4-5 bullet points in my research notebook to summarize what the paper is about, anything particularly interesting or useful to know, helpful figure numbers, etc. If I later come across a situation where I want to know what exactly the paper does, I can read the paper in more detail, which takes longer…

In my advisor's group, we did a helpful exercise, where we each bring a couple of papers to the meeting. We spend the first 30 minutes reading the papers silently (15 minutes each) and then we summarize/discuss the ideas in the remaining time for everyone else. It's a great way to make it through a lot of papers quickly