Questions

1. What's the main difference between soft and hard clustering?
2. Fuzzy c-means can easily deal with non-convex clusters. Is this statement correct?
3. Which is the main assumption of a Gaussian mixture?
4. Suppose that two models achieve the same optimal log-likelihood; however, the first one has an AIC that is double the second one. What does this mean?
5. Considering the previous question, which model would we prefer?
6. Why would we want to employ the Dirichlet distribution as the prior for the weights of a Bayesian Gaussian mixture?
7. Suppose that we have a dataset containing 1,000 labeled samples, whose values have been certified by an expert. We collect 5,000 samples from the same sources, but we don't want to pay for extra labeling. What can we do in order to incorporate them into our model?