7. Suppose we have observable variable X and that we aim to nd the value of pa rameter 9 to mam'nnze the likelihood (8;X) = p(X;9), where X is the data matrix X = {x1 , x2, ...,x,,}. Suppose we can introduce latent variable Z in the model; we have 37099) = fMX, 2:9)0'2- With one paragraph, explain how the EM algorithm can be derived from the perspective of maximizing the socalled Evidence Lower Bound (ELBO) l(q, 0) : f q(Z) log Mdz 4(2) with an alternating optimization procedure, where q(Z) is a probability distribution. (L)