Revolutionary preprint (part of ongoing Huawei Research work) has been published in ArXiv https://arxiv.org/abs/2106.14587v1 . It involves category theory, homologies, topoi, stacks, sheaves. Generally I am can handle this kind of math, no problem, but paper is a bit vague sometimes, it is full of revolutionary ideas that should be elaborated in the exactness that is required for scientific publication. But anyway, this work is too important to be delayed.

I therefore, have many questions about this paper. And I don't know where to put those questions - in Math or AI sites.

I have put my first question in math site: https://math.stackexchange.com/questions/4189699/topos-and-stacks-of-neural-networks-how-to-understand-sentence-values-in-t But I am afraid whether this is approriate. Therefor I just wanted to know the guidance from the community? Are AI people ready to answer questions about this kind of deep learning theory? This is more general problem anyway - is Stackexchange appropriate for cross-disciplinary science?


1 Answer 1


In theory, questions about any topic of a paper in machine learning, deep learning or neural networks are on-topic here, so I think your questions would be on-topic here, so you could try to ask them here, but ultimately it's your choice.

The problem is: do we really have the people capable of answering questions on such advanced topics? I do not think so, at least, I don't think that the regular users have the required knowledge, but I could be wrong and maybe some new/irregular user would be able to answer your questions. Category theory is really an abstract subfield of math, which shouldn't be taught in any AI curriculum/programme. Maybe if you study functional programming with Haskell you will learn about some of the related concepts, but, in AI, I don't see many potential applications of category theory, and that paper may be a good example or not of a potential application.

In any case, typically, you will not find many people capable of answering questions related to research papers, whether you ask them here or any other site. To understand a research paper, even for people involved in the field, requires some time and effort. In particular, in machine learning, it's not like reading a regular book or a tutorial: it's often a condensed reading that contains many technical/mathematical details that are not easily understandable, unless you really have a solid knowledge of the topic and prerequisites.

By the way, I read your post on Math SE, and I don't see how your initial claims could be close to the truth. Why/how would category theory ever be able to overcome the current limitations of deep learning? Without reading that paper, I don't really see why/how, but I only have a vague idea of what category theory is.


You must log in to answer this question.

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