Conjugate Gradient Method(共轭梯度法)
Goal: Make E(theta+y) smallest
It selects a matrix E''(theta), and all the search directions are mutually conjugate on E''(theta).
For Scaled Conjugate Gradient Method, if E''(theta) is not positive definite, add a scale onto it and make it positive definite.
see:http://class.htu.cn/nla/chat5/sect5_1.htm