Kernel Trick

• The Kernel Trick is to project data to a higher dimension.
• Typically, you use the trick when there is not a good linear separator in you current dimension scheme.
• 
• In this case, the kernel function translates the x and y coordinates to z coordinates.
• It's going to be something like 7-(distance from 0,0).
• The general problem is that you don't know which kernel to use.
• I just looked for a picture that involved a circle, because it's pretty easy to explain.
• The problem is made even more difficult by outliers.
• You could use the training data to make a kernel.