Why is SOM not being used?

Question

What SOM does sounds good and very useful on a paper (putting similar individuals close together, nonlinear version of PCA) for visualization and for dimensional reduction. Also there are whole chapters about it in ML books (e.g. Marsland), but I newer saw it being used in any serious way, also it was not merged in scikit-learn for the same reason.

If it sounds so good, why it is not? And if it so terrible, why is it being mentioned in ML courses, instead of more useful approaches?

as to teaching, perhaps more traditional perceptrons are a simpler concept? — a different ben, Jul 14 '15 at 06:25

score 2 · Accepted Answer · answered Jul 14 '15 at 06:32

2

What do you mean by "serious way"? A timeline of slightly outdated collections of (now over 10000) published (and I guess peer-reviewed) papers based on the Self-Organizing Map are listed at the website of Kohonen's alma mater: Bibliography of SOM papers.

One of the links on that website contains the text:

Many of the papers on SOM analyze the method or present variants or generalizations of it. Most of the papers, however, apply the method or its variants in fields ranging from engineering (including image and signal processing and recognition, telecommunications, process monitoring and control, and robotics) and natural sciences to medicine, humanities, economics and mathematics.

answered Jul 14 '15 at 06:32

a different ben

577
4
14

By "serious way" I mean papers that use it for reasons different than a comparison of various methods. So they used SOM to analyze/visualize data, not to show that it is possible to do it with SOM. Some of the papers in the link you posted are like that, so I might be just biased, although the papers are old. But it's still used significantly less than PCA, SVD, ICA... – rep_ho Jul 15 '15 at 14:16
I'll send you a copy of my paper when it's done! – a different ben Jul 16 '15 at 23:30

Why is SOM not being used?

1 Answers1