Hacker News Reader

I see a ton of papers about kernel-based stuff, but zero real applications. Feels like that whole subfield came and went before I got into ml.

Any time you're fitting a modeled distribution to an empirical one you can use MMD, for instance.

You have never used Gaussian processes?

SVMs are incredibly performant and is one of the most easy to use, low hanging fruit if you have a practical ML problem.