I'm a bit confused about the E step on the EM derivation for a mixture of multinomials. I don't think I'm all the way there, but I'm confused about how to proceed. Is the E step, solving for tau? Or is it something else?

In general, what are the steps? (This is what I think the steps are).

  1. Setup the log likelihood.
  2. Realize the sum is inside log so we want to simplify some other way.
  3. Realize that we need a latent z variable so we have to factor that in.
  4. Multiply by p(z,x|u)/p(z,x|u) (effectively 1, so not changing the value).
  5. Rearrange so we can use Jensen's inequality to move the sum out of the log.
  6. .... ????? Is this result the e step?

Below is where I am so far.

enter image description here

  • I am confused is this a derivation that you have come up with or is it something you are trying to understand? – Chinny84 Nov 17 '14 at 14:02
  • I came up with this (based on examples from Bishop's book). – William Falcon Nov 17 '14 at 15:59

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.