Hidden markov networks

Question 1

What problems could be solved with pattern recognition through time?

Answer 1

• Speech and writing recognition.

- Most human activities through time like playing basketball, driving

Question 2

What is dynamic time warping

Answer 2

• It is used so analyze how similar two signals are. Time warping is related with the idea that two signals have similar pattern but one happens in a larger time span. For example saying “Cow” fast and slow
• This method’s flexibility allows two time series that are similar but locally out of phase to align in a non-linear manner.
• Frecuency deltas are used
• The euclidian distance is calculated as the square root of the sum of the squares of the differences
• Do a graph where the two signals are in each axis. Then try to construct something close to the diagonal and estimate the euclidian distance

Question 3

What are sakoe chiba bounds?

Answer 3

• Putting some bounds from the diagonal of the dynamic time warping to dont allow big deviations

Question 4

hidden markov model

Answer 4

• ## Useful tool for looking at pattern recognition through time. Can be used for speech, writing or gestures

Question 5

How to train a HMM

Answer 5

1) Divide the number of timeframes in groups with similar number of members. The number of groups is the same as the number of states
2) Calculate the transition probabilities using the number of time frames in each state
3) Calculate the output probabilities
4) Try to reorganize the groups, assigning members to the group closest to its mean
5) Recalculate

Question 6

Baum welch

Answer 6

• Each time frame contributes to each state proportionally to the probability of that timeframe of belonging to that state

Question 7

Stochastic beam search

Answer 7

It is used to only consider the paths in HMM with the highest probability, to not consider all possible paths