49

Proposal: Machine Learning

Machine Learning is a valid and avid topic on CrossValidated. So what's the problem?

EDIT: While I can't see any justified point why this can't live happily within CV's [machine-learning] tag, I change this into a close request.

  • Do any of the other signal processing like proposals also overlap with CV? – Ivo Flipse Jan 28 '11 at 15:20
15

Machine learning is explicitly in the charter, but I wonder if all machine learning topics are a good fit for Cross Validated: there do not appear to be any questions relating to inductive logic programming, and one answer that mentions abductive reasoning in passing. The kind of machine learning well-represented on CrossValidated deals with statistical and connectionist methods.

I'd leave the proposal be for now, though it may turn out that the Cognitive Science proposal, which addressed the scientific part of the failed artificial intelligence proposal, will be a superset of this one.

  • 1
    I believe there is a place ILP on CV; and if not it could also exist on TCS. – mbq Jan 29 '11 at 17:17
  • 8
    Just because the questions aren't there, doesn't mean nobody should try and ask them! – Ivo Flipse Jan 29 '11 at 17:30
  • 4
    as says @Ivo, I think it should be tried to "seed" more ML on CV. Statistics and ML have a lot in common - and both communities really should speak more with each-other. Moreover, with two sites people will wonder, "should I ask that on CV or on ML?"; and others will comment: "hey, this is a question for the other site". Above all, I think fragmentation is unproductive when you're dealing with (very) broadly the same topic (which in this case I like to call "trying to find how Y relates to X"). Thanks god there is not a C++ stackoverflow, a java one, a perl one, etc. – Arthur Oct 15 '11 at 14:35
16

Cross Validated seems to be interested in only a subset of machine learning science, particularly the established machine learning methods that apply to data analysis in the narrowest sense. You will not find there many (if any) topics on machine learning as it relates to cybernetics (for example, systems control and robot control) or computational neuroscience. You will find only meager treatment of the subject of neural networks, which, apparently, is not a contemporaneously powerful approach in statistics, and yet is obviously important for other reasons.

For an example of the comprehensiveness with which Cross Validated treats machine learning, see this response to the question Backpropagation algorithm, which is possibly one of the most significant algorithms in the history of machine learning.

  • 7
    Ok, you seem to understand ML in broader sense; but judging from example Qs, this proposal is dealing neither with control theory (which IMO is for robotics) nor with this computational neuroscience (which was on topic on recently dead AI). About NNs on CV... sorry, but this is really an obsolete topic in data science; we tried, it was stimulating, but there are much better methods now. – mbq May 29 '11 at 22:03
  • 6
    @mbq Whether you consider neural networks obsolete or not, it is certainly true that a) many people do not share your belief and b) many people still use them. That a particular community (CV) is apparently not interested in them while others are interested in them is a legitimate argument in favor of a separate site. – Michael McGowan May 30 '11 at 19:28
  • 3
    @Michael So make a Neural Networks proposal ;-) It may grow or fail, I don't mind -- but there won't be an ambiguity between two sites. Or simply come to CV and enlighten our dark minds with splendid NN solutions to some problems posted there; no-one will ban you for that, and we may learn something exciting. – mbq May 31 '11 at 18:55
  • 3
    @mbq I'm not advocating a NN specific site, nor am I really a huge fan of NNs. I do think there is an issue though with an apparent lack of transitivity in scopes. You argue ML is on-topic at CV. @Matt Munson and I argue NNs are on-topic in a hypothetical ML site. I think neither position is contested. But it is apparently a point of contention as to whether NNs are on-topic at CV. Then again, perhaps there is no real issue to speak of: there are a couple dozen NN posts on CV as is. – Michael McGowan May 31 '11 at 19:35
  • @mbq I use NNs as the example because I have some familiarity with them, and thus have seen that they are not well covered at CV. However, the point of the example is that CV may only address the subset of machine learning science that relates to data science. However, I have no desire to see that ML discourse withdrawn from CV. I think the obvious reality is that SE (and academia in general) artificially delineates discrete domains of knowledge, when the reality is that concepts are much more intermingled. – Matt Munson Jun 6 '11 at 4:50
  • In other words, it would be cool if every stats related ML questions could show up on both CV and a tentative ML site. But, failing that, I think there remains a lot of discourse regarding ML that is not occurring, and could occur on an ML specific site. However, I appreciate the invitation to participate at CV, especially as it appears that the ML site may not reach fruition. Unfortunately, my interest in ML (for now) is primarily as a student, as my knowledge is very limited. – Matt Munson Jun 6 '11 at 4:59
  • 1
    The referred to response (#3443) to Backpropagation algorithm seems to have been deleted. – WikiSpeedia hang-around Aug 3 '11 at 9:01
2

Please see my post on meta https://meta.stackoverflow.com/questions/112747/the-living-and-the-dead-proposal-for-an-artificial-intelligence-robotics-s regarding this very subject.

I have already posted a notification on aiqus.com to try to rallye the troops. I have also attempted to contact the professors of the course to see if they can help leverage their audience.

Unfirtunately I didn't realize that I should have started asking questions here before forging ahead.

Uniquely, perhaps I am attempting to fuse the AI and robotics communities together as both have failed individually.

Please see the new proposal http://area51.stackexchange.com/proposals/36600/artifical-intelligence-robotics

0

Machine Learning uses a number of techniques or algorithms, such as regression, NNs, principle component analysis and SVMs. At its core it is a separate field, with unique perspectives on problems like Natural language processing, Bioinformatics etc. Those fields can be approached from an AI perspective (often based on hard-coded reasoning principles), or in a purely statistical way (visually or classically) for example. To reiterate, ML uses a particular breed of techniques (often learning by example or by like characteristics, while it cannot be defined as these techniques in disjunction.

  • 1
    I am not saying that ML is a part of statistics, only that there is already a SE site on which ML is on topic. It has confusing URL (stats.stackexchange.com), I admit, and that's why we prefer to call it CrossValidated. – mbq Jan 11 '12 at 17:01

You must log in to answer this question.

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