Support Vector Machines

• I've not used SVMs before, but they're currently flavour of the month.
• Instead of finding a line that separates the classes, find the line that separates them with the best.
• Get the support vectors; these are the points that are nearest to the other class, and if it's linearly seperable these are the points on a line.
• Get the line between these two lines.
• There's lots of cool math to help here.
• If the classes are not linearly separable, project the data onto a higher dimensional space. That's the kernel trick.