Mixture Modelling

Generally speaking, the likelihood in the Gaussian Mixture Model is non-convex and we should consider multiple restarts of the EM algorithm to find a good minimum

• True

• False

The Bayesian Information Criterion (BIC) chooses a model which has an optimal balance of....?

• variance and interpretability

• fit to the data and parsimony

• computation time and bias

Which of the following is common to both K-means clustering and Gaussian Mixture models

• The technique allows us to express uncertainty in the cluster assignment

• The optimal fit is obtained by maximising the likelihood

• The clusters obtained are all of equal size

Latent Class Analysis requires the data to be binary.

• True

• False