A new Machine Learning course

I have been learning about machine learning these days, and a good course I found is from 极客时间 机器学习 40 讲 https://time.geekbang.org/column/article/9762. The first three articles are about frequentist vs bayesian.

• In short, frequentist views parameters as fixed, but the data is limited in size, so we need to estimate the parameter using noisy data.

  • The disadvantage of frequentist, is that when the data size is too small, we cannot estimate the parameters reliably.

• In contrast, bayesian sees data as given, and the parameters follow a certain distribution, so we need to estimate the random variable of parameter from the fixed data.

  • The disadvantage of bayesian, is that the prior of parameter is hard to obtain, and it is subjective. We can use uninformative prior, such as a uniform distribution.