ABSTRACT
In the previous chapter we introduced the key concepts required to adopt a Bayesian approach to machine learning. Within the Bayesian framework, all unknown quantities are treated as random variables. Each parameter is described by a distribution rather than an individual value. Uncertainty in our parameter estimates is naturally channeled into any predictions we make. We saw two examples of prior and likelihood combinations that were conjugate, meaning that the posterior would be of the same form as the prior and could be computed analytically. Examples where we can justify the choice of a conjugate prior and likelihood combination are rare. In the remainder, we cannot compute the posterior and must resort to approximations. In this chapter, we will introduce three such approximation techniques.
