coursera Andrew Ng 机器学习第七周笔记 支持向量机

Support Vector Machines

optimization Objective

Large Margin Intuition

Kernels

Using an SVM

choice of parameter C

choice of kernel(similarity function)

if Gaussian Kernel, need to choose sigma^2

Not all similarity functions make valid kernels(need to satisfy technical condition called "Mercer's Theorem" to make sure SVM packages' optimizations run correctly, and do not diverge).

Polynomial kernel:
More esoteric: String kernel, chi-square kernel, histogram intersection kernel

multi-class calssification

K SVMs, one to distinguish one from the rest.