Hidden Markov Process

In a Markov model, each State represents an event that can be observed. Therefore, a Markov model is sometimes called a Markov Model (vmm), which limits adaptability in a certain Chengdu airlines. In the Hidden Markov Model (vmm), we do not know the sequence of States passed by the model, but only the probability function of the State. That is to say, the observed time is a random function of the State. Therefore, modifying a model is a double random process. (Statistical Natural Language Processing (version 2 ))

(If you do not know the Markov process, please first take a look at the Markov process)

The following uses a bag in the dark room as an example. Here I will briefly describe that there is a dark room with 10 bags containing different numbers and different colors of balls, for example, 3 red and 1 green are installed in bag 1. Bag 2 is green and black.

Then, if someone is in the room and can't see it, go to the bag and touch one, and hand it to the people outside. The ball seen by outsiders is called the output sequence (O1, o2... ot) (also called observation sequence), which may be red, green, black, red .. this is the case.

The above process is a hidden Markov process. Why is it called "hidden? The reason is that the state (the bag in the previous example) is unknown to the outside, and the symbol (output sequence, the ball in the previous example) is directly visible to the outside. Let's look back at the Markov process. Every State (or sequence) is fixed and one-dimensional. The process in the above example is a two-dimensional Markov, so it is called hidden Markov.

 

 

Q: Why does it mean "Observation sequence" instead of "state "?

In order to take advantage of Markov properties as much as possible, this modeling seems to simplify the problem. If you have to roll back, isn't it too complicated?

Unfinished