(From yesterday to now, I have learning EM for 4 times, and each time I will gain some new ideas and correct some misunderstandings. So cool!)
Finally knows that the P(x(i),z(i);θ) means the joint distribution of x and z, parametered by θ.
Only if the taget data's distribution belongs to exponential family, we could use EM to estimate the distribution of the input datas.