Applicable scenarios: Estimate model parameters when there are unmeasured variables .
EM algorithm:
input: observation data Y, which is observation data Z, joint distribution P(Y, Z|θ), conditional distribution P(Z|Y, θ)
output: model parameter θ
step:
(1) Select the initial value of the parameter to iterate
(2) Step E: Seek expectations
(3) M step: maximize the current θ
(4) Repeat (2) (3) until the algorithm converges
Example: Different shapes of peas.