kernel and kernel-trick are two different tags on CV, with the first one reserved for kernel smoothing and density estimating, and the second one for kernel trick in maching learning (kernel SVM, kernel PCA, etc.). I noticed that a lot of people confuse these tags, mostly by using kernel instead of more specific kernel-trick.
This is bad, because e.g. questions about kernel PCA are currently divided (approximately evenly) between these two tags. I see two possible solutions:
The questions confusingly tagged should be re-tagged.
Personally, I think that the tags are separated quite well and would therefore prefer the second option. Moreover, I was going to go ahead and re-tag all questions about kernel PCA to kernel-trick (and possibly some other questions about SVMs, if I come across any). This would, however, bump all these questions to the front page, and if I re-tag them in one sitting, then there will be lots of questions on the same topic edited by me appearing on the front page practically one after another.
Would this be okay?
[Of course, I can also make some other improving edits and maybe even answer some of these questions along the way, to make this less of a book-keeping activity. But still.]
There are currently 68 questions tagged with kernel and svm and 7 questions tagged with kernel and pca. Whereas I was originally referring only to questions on kernel PCA, the problem seems to be much more severe in the SVM topic. For comparison, there are only 16 questions tagged with kernel-trick and svm.
Turns out, this issue was discussed two years ago: What does the kernel tag refer to? where @mbq and @whuber argued that kernel-trick is unnecessary. I think I am happy to accept this point of view, but that would mean that kernel-trick should be merged with kernel.