2
$\begingroup$

In a statistical model with parameters $\theta$ and unobserved laten variables $Z$, the model likelihood is

$$L(\theta;X)=Pr(X|\theta)=\sum_ZPr(X,Z|\theta)$$

The standard way to estimate $\theta$ is to maximize this likelihood (e.g. using EM or so). However if this estimation is computational intractable, can I maximize $Pr(X,Z|\theta)$ w.r.t both $Z$ and $\theta$, and use this estimated $\theta$ as the model parameter? Does this method have a name? What is the main drawback or problem with this method?

$\endgroup$
1
$\begingroup$

The problem with that approach is that every time you add an observation of $X$, you also add an "observation" of $Z$. The space in which you want to find the optimal values of $Z$ and $\theta$, thus grows very large. Instead of making the problem easier, I think you make it more complicated by adding $Z$.

If the optimization problem is intractable, you might consider using a smaller (or simpler) $\theta$. Your question does not give any details about the distribution of $X$ or about $\theta$, therefore I cannot give more concrete solutions.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.