Earlier quoted context omitted.
It's a practical (especially for small to medium sized data sets), effective way of doing machine learning that has a highly sound theory behind it and doesn't require arcane tricks that are rarely or never written down, unlike old-style artificial neural networks. The interest in kernel methods is proof that there's nothing more practical than a good theory. :)
At the same time, I think it's astounding how often k-nearest neighbors (a method considerably simpler than SVMs) is either the best or second-best classifier for a dataset. I'm not trying to pit SVMs vs. neural networks...rather I'm saying most methods of machine learning (neural nets, SVMs, gaussian mixture, kNN, what have you) can be effective and none stand out above the rest.
Re: Major research breakthroughs in the last 10-15 years
#61Woops...forgot the squeeze an "often" before the "either"...