12 - HMMs

Which of the following descriptions is closest to Hidden-Markov-Models (HMMs)?

• K-means + online PCA on temporal data.

• Paiswise clustering on temporal data.

• Mixture models on temporal data.

• ICA + PCA for whitening on temporal data.

What is the forward-backward procedure used for?

• Speed up computations.

• Finding a closed-form solution to maximizing the likelihood.

How do we compute the parameters of the HMM?

• we have a closed-form solution.

• Expectation-Maximization (EM).

What is the Baum-Welch Algorithm about?

• Combining fwd-bwd + EM to estimate the parameters of the HMM.

• Compute the posterior distribution.

• Decorrelating the data.

• Normalization.

What does the transition model describe?

• The initial distributions.

• The initial distributions + transitions probabilities.

• The transition probabilities + emission probabilities.

• The emission probabilities.

What is the common definition for the emission probabilities?

• A Gaussian distribution.

• A mixture model.

• A kernel density estimate for the x's.

What can we extract from this diagram?

• The dependencies between the variables.

• The size of the parameter space in the HMM.

• This HMM is designed to only predict observations.

• The conditional distributions.

Which distributions are the result of the Markov properties in the HMM?