Question

Mixture Models and Digit Classification

Assume that we have 100,000 black-and-white images of size 26*26 pixels that are the result of scans of hand-written digits between 0 and 9.

We can apply mixture models to effectively train a classifier based on clustering using the EM algorithm applied to the dataset.

Identify the following parameters (according to notation developed in the lecture, assuming that we use all the data for training):

Answer 1

Identify the following parameters (according to notation developed in the lecture, assuming that we use all the data for training):

K=
n=
d=