《Gaussian Processes for Machine Learning》
考虑下面这个经典的线性回归的例子，目标函数是
f(\boldsymbol x) = \boldsymbol x^T\boldsymbol w
观测到的值为真实值+高斯噪声\epsilon ~ \mathcal N (0, \sigma_n^2)
y = f(\boldsymbol x) + \epsilon
