It's been a few years, but I'll take a shot. 1. SVMs are useful for categorizing items. 2. SVMs can only categorize on 'linearly separable' sets. That is, if

It's been a few years, but I'll take a shot. 1. SVMs are useful for categorizing items. 2. SVMs can only categorize on 'linearly separable' sets. That is, if you were to plot the items from the two sets on a plane, it's possible to distinguish the two sets by drawing a line on the graph and everything above the line is in one set, while everything below is in the other. 3. The kernel trick is adding dimensions to a problem, so even though the problem isn't linearly separable in two dimensions, it might be in 3 by setting the right values in your extra dimension.