Frequencist vs Bayesian viewpoint of language models

TODO: explain the different philosophies

the parameter in bayesian is considered random or has uncertainity like the ${\theta }_k$ (prior) parameter for (but how does is encode uncertainity ? what does the paramter mean exactly) and that we remove unvertainity by learning using fixed data and getting the posterior $P(\theta |D)$

The data is fixed in bayesain viewpoint, lije this is what I got so I use it to increase my belief

TODO Few points to elaborate more:

1. MLE vs MAP (Maximum posterior estimation) look for the paramter that gived tha maximum estimate

2. How does Bayesian doo model averaging (Bishop)

3. Dropout is a form of model averaging or model averaging in gereal $P(y|x,D)=\int p(y|x,\theta )p(\theta |D)d\theta$ And that how does model average does not collapse to a singe point estimate weight, it averages on all of them

4. MAP has the regulizer in it naturally, there is still alot to cover