2014HS: 33065 Applications of Artificial Intelligence

Hi, revising for the exam i came up with some questions. It can be useful for everyone if somene could answer them =) Here they are:

1. Haming windows: Better Wide or Narrow ones? (Resolution vs Detection) I recall during the lesson we discussed a minumum size for not losing frequency 300-3k (Human speech) am I right?

2. N-GRAM model (slides 4, 19/45) : GSF and WIP, how are they computed? What is the interpretation? The lower the better ? Or is the formula above the interpretation ?

3. What is the real difference between HMM and ANN ? I looked also in the web but they all give statistical based answers, I would like to know when do use one and when to use another. Furthermore, i read that HMM are a special case of ANN, how is this the case?

4. PCA, what is its purpose informally ? Is it used only for feature selection ? How ?

5. RNN vs NN ? Is just the memory possibility ? What does it really mean that a NN has a "memory" ?