最近刚刚看完Ng的machine learning，想再加一点料，于是拿起PRML，开始啃。。。
第一章：
Polynomial Curve Fitting:
regression, error function, RMS(root-mean-square),overfitting
Probability Theory:
many distribution…..
- frequentist: think parameter exist, we just need to find it
- Bayesian: think paremter is random variable, we need to take its distribution into account
- frequentist and Bayesian diff in prior of parameter
- curve fitting(probability aspect): model P(x|w) –> error function, model P(w|x) by Bayesian –> RMS
- Bayesian curve fitting: integrate all w
Model selection:(pass)
Curse of Dimensionality:(pass)
Decision Theory:
- Minimize misclassification rate, minimize expected loss, reject option(set threshold)
- three type:generative model(model p(x|y)), discriminative model, no probability
Information Theory:(pass)
import views:
1. frequentist and Bayesian
2. regulation
3. regression on probability view and non-probability view
4. generative model, discriminative model, no probability