Bayesian thinking treats probability as a degree of belief that you update as evidence arrives. You start with a prior, combine it with the likelihood of the data, and get a posterior. This view is behind many AI methods, from spam filters to models that report how uncertain they are.
The lessons first separate probability from likelihood and show how maximum likelihood fits a model to data. Next come the ideas of priors, posteriors and conjugate priors. They finish with Naive Bayes, a simple classifier built directly on Bayes' theorem.
Watch the lessons in order, then continue on to Information Theory & Loss Functions.
Explains the difference between probability and likelihood, two ideas that are easy to mix up.
Video 5minShows how maximum likelihood finds the model parameters that best fit the data.
Video 6minShows how the prior and the data, through the likelihood, shape the posterior.
Video 8minCovers updating a model with new data and why conjugate priors make that easy.
Video 17minExtends Naive Bayes to continuous data using the normal distribution.
Video 9min