4
$\begingroup$

In the vast world of deep learning, there are countless models each having its own unique set of properties. Let's say we have a model M with a specific property P.

Now, if I have a question about model M exploiting property P, and I'm interested in any existing research or studies that delve into this particular aspect, would it be considered appropriate to ask here?

In other words, are questions framed like, "Are there any existing studies or research focusing on leveraging property P of model M?" acceptable within the guidelines of this forum?

I am keen on learning from the community and expanding my understanding in this field. I believe such questions could lead to fruitful discussions and knowledge sharing, but I wanted to make sure they align with the forum rules.

$\endgroup$
2
  • 2
    $\begingroup$ A formal literature search would be quite different to someone finding a single online article, and there will be things in-between, such as finding one matching published academic paper. Could you clarify which end of this spectrum you would like to see Q&A about on the main site? $\endgroup$ Commented Aug 3, 2023 at 7:23
  • $\begingroup$ @NeilSlater I am asking for question like "Are there any studies examining the impact of image order on the classification capabilities of Convolutional Neural Networks (CNNs)?" so finding recent published academic paper type of task. $\endgroup$
    – satya
    Commented Aug 3, 2023 at 8:12

2 Answers 2

2
$\begingroup$

"Find me an external resource" type questions are often considered off-topic across multiple Stack Exchange sites. Although some sites do encourage it, based on their remit.

There are a few issues I can think of:

  • It's hard to prove a negative, so you will very rarely get "there is no such study" answers, even though that could be the correct answer. Instead the question will remain unanswered, which is generally a statistic SE sites like to minimise.
  • A correct answer that simply links to the external resource would be considered low quality.
  • Published academic works can be behind paywalls.
  • The results of a search for external resources can change over time, whilst an answer to a specific technical question may become more or less relevant, it should not expire as easily.

As a question asker, you have a couple of ways to adjust your question to achieve similar results, but you have to shift your expectations away from obtaining links to papers to read up on a subject.

You can ask any technical question, and suggest at the end "Any links to academic papers on the subject would be appreciated", or similar hint about what you'd like to see.

You can also phrase what you are looking for as a more direct question about the subject e.g. your example "Are there any studies examining the impact of image order on the classification capabilities of Convolutional Neural Networks (CNNs)?" becomes "Does image order affect classification performance of CNNs?"

The example question is interesting, in that the answer would not refer to any studies of ordering image input to CNNs. That's because much earlier work showed that randomising dataset order for mini-batches was much better for performance. Chances are that no-one will have published a formal study of providing training data in a fixed sequential fashion for a CNN. So by asking for that, you cut yourself off from a useful technical answer of why that is a bad idea.


In summary, I think Stack Exchange is not the correct service to collate results of literature searches. Answers can contain links to supporting literature as appropriate, but the original question should not be "please find me a paper about X".

$\endgroup$
1
$\begingroup$

In other words, are questions framed like, "Are there any existing studies or research focusing on leveraging property P of model M?" acceptable within the guidelines of this forum?

That's acceptable to me. A good amount of research time is spent on finding relevant research papers, and therefore I would a correct answer that simply links to a relevant research paper as high quality. Good AI research is rarely behind paywalls, and even if when it is, it can still be accessed, e.g. via universities/library subscription or Sci-Hub, and it's not OP's fault when paper authors decide to support the paywall industry.

Examples of reference-only answers:

All of them have a strictly positive score and no downvotes. I didn't cherry-pick, I just looked at my latest answers.

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .