Gaussian Mixture Models

Question 1

What is the EM algorthm?

Answer 1

1. Start with random gaussians
2. (E-step) Compute posterior P(c|x) for each point
3. (M-step) Adjust gaussians (mean/variance) to better fit points
4. Repeat until convergence

Question 2

Bayes Inf. Criterion

Answer 2

max_p { L - 0.5 * p log n }

Where L is the likelyhood, and p is the number of parameters

Question 3

Akaike Inf. Criterion

Answer 3

min_p { 2p - L }

Question 4

Whats the general form of a gaussian?

Answer 4