- Bayesians probabilistic, frequentists only use past beliefs.
- Start with a belief, your prior. Obtain data and use it to update the prior, resulting in the posterior.
- Bayesian ML: use Bayes rule to infer model parameters from data
Inference: how you learn parameters of your model. Model is
separate from how you train it!
- Different from deep learning, where all training mechanisms use SGD. Bayesian inference has more different methods
- MCMC and variational inference
- Two flavors of Bayesian methods: statistical modeling and probabilistic ML