9 - Hidden Markov Models

Question 1

Filtering

Answer 1

Computing the BELIEF STATE (posterior) over the most recent state given the evidences up to the present

Question 2

Alternative Name for Filtering

Answer 2

State Estimation

Question 3

Smoothing

Answer 3

Computing distribution over PAST states given the evidences up to the present

Question 4

Prediction

Answer 4

Computing the POSTERIOR over the FUTURE STATE given the evidences up to the present

Question 5

Most-Likelihood Estimation

Answer 5

Recursive Viterbi Algorithm

Question 6

How are tasks in HMM computed?

Answer 6

Recursively

Question 7

Hidden Markov models have a ________.

Answer 7

Hidden Markov models have a SINGLE DISCRETE STATE VARIABLE.

Question 8

HMMs are e.g. used in ________.

Answer 8

SPEECH RECOGNITION - STOCK MARKET ANALYSIS - DROWSINESS DETECTION

Question 9

Why is it not necessary to normalize the intermediate results of the Most Likely Sequence?

Answer 9

We are only interested in the best sequence and not in the absolute probabilities.

Question 10

When is it not possible to precisely predict the state?

Answer 10

No accurate predictions for a LONG time horizon

Question 11

Difference between Filtering and Prediction

Answer 11

Prediction: posterior over the future state Filtering: posterior over the most recent state

Question 12

A Markov process is a ________ process that has the ________.

Answer 12

A Markov process is a STOCHASTIC process that has the MARKOV PROPERTY.