Why do we really need it?

Question

OK, so, after reading some discussion here I think everyone should calm down and think about this proposal with cool head. I've jumped in and was pretty excited about it at first, but now I'm becoming more and more skeptical. I think we should consider the following reasons before actually committing to the new site:

1. I don't see a single question in proposed list which is not appropriate for Cross Validated. So, there exists a site where all these questions could be asked and where all the same people (plus more) are present. So it doesn't give any improvements on coverage of topics or users...

2. Cross Validated is not hugely popular. It is just OK. And machine learning is 4th most popular tag there. That means we are going to split out a big chunk of CV, which will make two semi-dead pages out of one OK page. This is considering people will actually move to ML, I feel like most will just keep visiting CV and ML will have an extremely low traffic. I'm not even starting on confusion about where to actually post the ML-related questions (given the already popular category on CV). What will happen is that most of ML guys will be registered on both and newcommers will be pretty confused.

I'm writing all this simply because I think whatever is done here should be done for the benefit of the community. The reasons for opening a new page would be if there is a big area of questions which are not covered in already existing SE page or if there is next to none overlap of this area with already existing site. Neither of these is true in this case. "We want to get apart from those linear regression guys" is not a valid reason.

Don't get me wrong, if it does go through I'm going to participate actively for sure, I just don't think it is going to improve anything. I'm passionate about machine learning and do understand when people are excited about it, but I'm also passionate about SE and in my opinion emotions are not what should be the motivation.

All in all, please think in terms of benefit to community first, and then vote. I'm happy to discuss further if people have different opinion.

Proposal: Machine Learning

Answer 1

There is already an ML tag in CV (205 followers, 1164 questions; total CV users ~17k) and one in CS beta (it's ranked 25 of 432, 30 followers, 84 questions; total CS users ~6k). Presumably CS will grow once it goes public and so will followers of the ML tag there. I agree with those here who say that splitting out ML will dilute the exposure. If I had a mathematically slanted ML question, I'd go to the CV board. If it was CS slanted, I'd go to the CS board. Regardless of how things are classified, there are already places to get those questions answered. The assertion that CV is "really ... a machine learning board" only supports the argument that creating a separate ML board is not in the best interest of the community because people already know where to go for answers. So,I think the best answer to the question "Why do we need it?" is that we don't.

Answer 2

I think that there is a good chance that an ML site would incorporate theoretical questions from MSE like this one or this one, and implementation questions like this one from CV.

I asked or answered all of the linked questions above, and it would be nice to see them answered in their own area, or given attention as such (I believe I got Tumbleweed for one of them). Mathematicians, for one, may be less interested in some of the shaky ground on which adaptive filters like LMS or other learning algorithms are based.

Answer 3

The real answer to your question is that CV is really (so I would argue) a machine learning board. It seems to be pretty consistently concerned with statistical inference and, more particularly, the methodology of practical statistical inference, which are the primary concern of ML scientists. It is not usually primarily concerned with things like asymptotics, convergence, guarantees, and other things that dyed-in-the-wool statisticians are usually concerned with.

But here's the problem. It is a good ML board, but it is not billed as a ML board. People don't think of it as an ML board, and they don't advertise it as an ML board.

The reason this is a problem is because, especially for lay people, it is not clear where to go to talk about ML. MetaOptimize? CV? Some Google+ group? Which place has the best community? Which is most likely to answer your questions? One of your questions is whether there's enough room in this town for all of us. That there are already so many medium-traffic communities indicates that, yes, there is. The question for me is: where have these communities failed, and how can we build a better one? Because, IMHO, in one way or another, they have indeed all failed.

Of the existing communities, I think CV is the best, but it is not clear that people would know to go there, and it is not clear that the community would want to be called an "ML" community, since statisticians -- which is what they would say they are, I think -- think of ML as a faddy renaming for vanilla inference.