In: Statistics and Probability
Explain the role of the kernel function in Support Vector Machines
Kernel :
It is nothing but a similarity function as the domain expert provide to machine learning algorithm. It splits out the two inputs how similar they are.
why we take kernels, talking about this one big reason is computing the feature vector corresponding to the kernel is really really hard when compared with the computing the kernel.
Example:
consider following image we have a toy problem separating the blue crosses from the circles on a plane. here in the left figure ellipse acts as a seperating surface. here kernel trick is transforming the data in to 3-d space would makes much easier to do.
a kernel is not only special for SVM and it is not restricted. This algorithm can be useful in anywhere having a an algorithm that works with dot products. Here the idea of kernel trick is good and better for SVM that makes good choice.