From my experiences with this site, I've noticed standard proofs in statistics and probability theory, generally on foundational, textbook topics, tend to get answered and dealt with extremely quickly.

I recently posted a question about the Vapnik-Chervonenkis inequality below, and after some digging , it turns out the question is not trivial in the sense of requiring detailed comparison of two hefty proofs.

What are the arguments in getting from the theorem of Vapnik & Chervonenkis (1971) to the common form seen in Devroye, Györfi & Lugosi (1996)?

Because I wasn't sure whether it appropriate to post at Cross Validated or Mathematics Stack Exchange, I placed a comment below it saying that I would migrate it myself to the latter if it wasn't appropriate. That comment received 2 upvotes, possibly indicating a degree of support for migration, but I am soliciting a more authoritative, explicit position from the community.


Are questions concerning proof in journal papers more suited for Cross Validated or for Mathematics Stackexchange? In particular, I mean proofs which are non-trivial that do not occur textbooks, but say in machine learning, theoretical statistics, and statistical learning theory journals.


On the one hand, my view is that the 'sanitised' world of abstract proof is much more in the spirit of Mathematics Stackexchange, rather than the 'messy' realism of data analysis and hence Cross Validated. This would suggest Mathematics Stackexchange for this kind of question.

On the other hand, it would seem that proofs in machine learning, theoretical statistics and statistical learning theory, while 'mainstream' to MSE in the sense of being proofs, are also esoteric in the use of specialised concepts, terminology, notation that are not within the realms of more traditional subfields in mathematics. In my opinion, the difficulty that this creates is that whilst I may be able to solicit assistance on parsing a proof on MSE, some of the proof steps may be motivated by specialised understanding that is more commonly found on Cross Validated.

As an example, how does one post a question on MSE, and reasonably expect someone there to assist in parsing a proof in the Annals of Statistics or the Journal of Machine Learning Research about theoretical guarantees on the AdaBoost algorithm if knowledge of its workings are in short supply? Clearly, proof is in the realm of Mathematics Stackexchange, but the requisite knowledge to understand its motivations is in greater supply on Cross Validated. Can someone advise?

  • 4
    $\begingroup$ The question is perfectly on-topic here. I'm not sure why your comment got upvotes -- there's no need to migrate it. The tricky thing about asking a detailed, specific question in a high-level topic like your question about V&C is that you need someone with similarly niche knowledge to answer. Means your audience is smaller -- but there's nothing wrong with the question. Offering a bounty might help draw attention. $\endgroup$
    – Sycorax Mod
    Aug 31 at 16:35
  • $\begingroup$ @Sycorax. Thank you for taking the time to provide with me input. How about if I replaced 'on-topic' i.e. not inviting closure, with a more slippery notion of likelihood of being answered? From my experience, I've found that more involved theoretical questions on important results in modern machine learning, theoretical statistics and statistical learning theory tend to go unanswered, possibly indicating a greater orientation on this site towards data analysis. As a moderator who's probably been on here for longer than I have, is that coherent with your experience on here? $\endgroup$
    – microhaus
    Aug 31 at 16:43
  • 1
    $\begingroup$ @Sycorax. Thanks, your edit answered my queries. It is really helpful for me because I had become hesitant to post further questions about the proof of the result here. $\endgroup$
    – microhaus
    Aug 31 at 16:49

You must log in to answer this question.

Browse other questions tagged .