Bayesian inference

Using Bayes’ Theorem to infer the parameters of unknown posterior distribution from data

Approach

• compute the maximum likelihood estimation 
• sampling (markov chain monte carlo methods)
• fitting (variational inference)

Try distribution with fewer unknown parameters if we can

• It’s expensive to run models with multiple parameters
• Underlying each parameter is assumptions. The more parameters we have, the more we need to express uncertainty.

Maximum likelihood estimation