EM in a Nutshell

Henry Kautz – CSE 473 – Intro to AI

Nov 24, 2003

Consider learning a naïve Bayes classifier using only unlabeled data:

Initialization: randomly assign numbers to P(C), P(A|C), P(B|C).

E-step: Compute P(C|A,B):

M-step: Re-compute maximum likelihood estimation of P(C), P(A|C), P(B|C):

Calculate log likelihood of data:

If still improving (increasing): repeat at E-step.

Otherwise, stop.