Reading Research papers

AI Discussions
1

I am beginning to attempt to read and comprehend research papers, and eventually hope to implement new ones I read. Whenever I give it a shot though, I come across so many terms I haven’t heard of before which I google, which leads me into a spiral of information and the cycle repeats and I make progress very slowly. Does anyone have any pointers, or is there a tutorial out there that walks through a paper explaining each word?

I find it easy to understand the general gist, but since I plan to write my own in the future(taking baby steps for now of course), I need to understand the terms used in literature.

Any other advice for a person planning to pursue research in this field would also be highly appreciated. Thanks!

1 Like
2

Reading research papers takes lots of practice until you learn to filter out jargon from actual message. I find it useful to try to structure the papers in the following way: abstract, Intro, Methods, Results/discussion, conclusions. Actual papers might vary but they all follow that overall structure. Always try to go from a high-level understanding to the nitty details, not the other way around!

  • start from the abstract, a one paragraph summary of the key new stuff and results. Try to put into your own words what the paper is about (e.g. a novel insight/model is presented that performs X% better on some benchmark dataset, OR a novel application of an existing model, etc.)
  • skim through the intro. Here’s were the state of the art is presented and credit is given to people who have thought about the problem before
  • methods: This is probably the most technical section, feel free to skip in a fist read. Sometimes the data sample, how it was collected/annotated etc. is presented here.
  • results & discussion: here’s where the meat of the paper is, what are the novel results? why are they relevant? what are the limitations? what’s the scope for the future, etc?
  • conclusion: this is usually a recap of the paper “In this paper we’ve presented a … that leads to …”, like an extended version of the abstract. Future scope can be also presented here.
1 Like
3

You can hear Prof Ng give his advice on how to read research papers in this video from his CS230 Stanford course.
That course could be considered as an extension to this specialization. All of the videos are worth watching

4 Likes
4

@ngbusca @GordonRobinson Thank you both of you for some excellent suggestions and resources, I genuinely really appreciate it!