This preview shows page 1. Sign up to view the full content.
Unformatted text preview: . 21 In addition to the graph structure, we use plates to denote repeated, inde
pendent draws from the same distribution. The plates are like the ‘for’ loops
in the pseudocode of how the data is generated in the text description above.
Coin Flip Example Part 7. Even simple models like the coin ﬂip model
can be represented graphically. The coin ﬂip model is: 5.3 Inference in hierarchical models Hierarchical models are useful because they allow us to model interactions
between the observed variables (in this case the words in each document)
through the use of latent (or hidden) variables.
What are the latent variables for LDA? Despite introducing a larger number of latent variables than we have observed
variables, because of their additional structure hierarchical models are gen
erally not as prone to overﬁtting as you might expect.
We are interested in infering the posterior distribution for the latent vari
ables. Let Z = {zi,j }i=1,...,m,j =1,...,ni , θ = {θi }i=1,...,m , φ = {φk }k=1,...,K , and
W = {wi,j }i=1,...,m,j =1,...,ni . Then, by Baye...
View
Full
Document
This note was uploaded on 03/24/2014 for the course MIT 15.097 taught by Professor Cynthiarudin during the Spring '12 term at MIT.
 Spring '12
 CynthiaRudin

Click to edit the document details