25

I am a CS undergrad student who is just getting started with research and reading papers. My mentor has asked me to read a few papers. While reading I noticed that every few lines there are references to previous papers being cited and the entire paper just builds on the previously described architectures and then adds something new to it.

In such cases, if I start reading the cited ones (and there are many mentioned over and over again), I am afraid I might get stuck into a loop of such repetitions and never get to the main idea. How should my approach be ?

My mentor wants me to thoroughly understand the ideas stated here because we are going to be implementing those soon albeit with modifications. I know of the 3-pass approach but that does not solve my particular problem.

Note that my particular doubt is not on how long should reading take. I can see many resources explaining that very well already. So I don't think this is a duplicate of them.

  • 1
    possible duplicate of How long should it take to read a paper? – EnergyNumbers May 5 '15 at 6:05
  • 9
    OP didn't ask how long. – 299792458 May 5 '15 at 6:06
  • 3
    That question does link to math.stackexchange.com/q/617625/26580 – Superbest May 5 '15 at 8:40
  • How you experience a paper depends also a lot on what's the author's take in a pedagogical sense. You will notice with time some papers explain things sequentially and lead you naturally to the main result (possibly at the expense of being lengthier) while other papers will skim on the details and (hopefully) use references while doing so. It's considerably harder for the novice in a field to read the latter type. – Miguel May 5 '15 at 16:48
24

What you are describing is pretty normal in academia. People take on old and existing work, and build upon it.

There are typically several reasons for citing prior studies, thus the impact of each individual citation might (and usually does) vary as well. I reckon this is all quite new to you. In your case if you can understand the work presented in the paper without diving into the citations, then just do that and ignore the references for the time being. You can always trace back to specific (read: not all) references later on to fill in any potential gaps in your understanding.

If you cannot grasp the why or the how, then my suggestion is to pay more attention to the introduction section, checking out the important (often recurring) references, and then re-reading the introduction. You should get a fairly good idea of what the paper is really about this way.

Another approach could be to try to discuss the paper with your supervisor or someone else who might be able to judge it better (any grad students or TAs?)

  • 16
    I would stress "recurring references". It's difficult to know where to look, but if there's a reference that gets cited 20 times, it is probably one worth reading. Maybe even one that is more important for you than the paper you have in hands. – yo' May 5 '15 at 7:41
  • @yo' good point, tried to stress it a bit more – posdef May 5 '15 at 9:40
5

Research papers build on previous work, so it can be hard to get started. In addition to the excelent suggestions already given you can also look for a "review article", or ask your mentor if he knows such an article. In many disciplines it is common that so every once in a while a (senior) researcher publishes an article that summarizes and reflects on the current state of art concerning a particular question. Such articles are very good for getting started in a discipline, as long as you keep in mind that such articles are not necessarily a neutral representation of the field.

  • Yes, and the "Annual Reviews in [subject]" are of reliably high quality – David LeBauer May 5 '15 at 15:10
  • Also, look for articles that wear the name "survey". – JaBe May 5 '15 at 15:53
3

There isn't anything surprising or unprecedented about references to earlier works, and the authors building on earlier models/structures. Additionally, not every article (in any field) is going to be pedagogic. While that could be due to the author writing style, it is not always possible to keep everything at a very basic level too. This is true, in particular for the stuff like letter publications, or rapid communications, where there are limits on manuscript length. Also, when there are existing pedagogic review articles, or otherwise detailed accounts which could pass off as pedagogic, authors generally tend to avoid a detailed presentation, simply cite them and get to their main point.

Since you say that

My mentor wants me to thoroughly understand the ideas stated here because we are going to be implementing those soon albeit with modifications.

it is clear that he/she wants you to understand the basics thoroughly, and not just be limited to the arguments presented in the one article you have at hand. So, if this article appears cryptic, you have no choice but to go through the cited ones for clarifications. Try to see if they are less cryptic, it is possible that the authors may have cited a review article of the sorts I mentioned above. That will certainly help. Based on that understanding, when you revisit your particular article at hand, you will understand it much better and better understand what it adds to the subject.

However, if that doesn't work out, and you still find the arguments too cryptic, consult your advisor and explain why you find it cryptic, which arguments are not transparent etc. It helps to carry along with you the attempts you made towards understanding it yourself, so that he/she knows you are not asking to be spoon-fed. Once he's convinced that you made an effort from your side, he will generally point you towards other articles which could clear up the specific point for you.

I mean, there is no other choice. There is no way you can gain anything out of this collaboration, or implementing anything, unless you are clear about what you are doing.

  • +1 I made an edit to your second to last paragraph. Hopefully, this is what you had in mind. – Mad Jack May 5 '15 at 13:49
1

The abstracts of those linked articles should be sufficient to tell you whether they are being cited as a major contribution to the work you are interested in, and abstracts should be written to be quick to read. With experience you will find that some referenced papers can only need skim-reading, others can be left to last based on the title, and a few will need detailed follow-up (though you have to learn when to stop). Groups of references will often not need full attention paid to all of them.

Here's a little example, from a hypothetical paper about producing a better model building on previous models:

I could guess that in "methods based on the xyz technique[17-23] have been found to consistently underestimate reality in this region", references 17-23 probably won't need reading (in detail); in "the abc method[42-45] is a good estimator at low input values, while the pqr method[46-58] is more applicable for high values" I'd expect to read some or most of references 42-56. If the author felt helpful or their mind worked this way, 42 and 46 would be good places to start, and in any case you should pick up an understanding without actively reading all the references in the range. If the paper carried on to say "previous attempts[63-79] to combine these two approaches have shown some success, however..." you'd probably have quite a bit of reading to do, as you're presumably interested in the combination as you're reading this paper.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.