hidden markov model

Post on 15-Apr-2017

553 Views

Category:

Science

1 Downloads

Preview:

Click to see full reader

TRANSCRIPT

Hidden Markov ModelNghia BuiNov 2016

Andrei Markov (1856-1922)

2

The weather problem

• I talked to Jane for days through telephone. Everyday she told me what she does, either “walk” or “shop” or “clean”, only one!

• I know, on a day, the weather in her city can be either “sunny” or “rainy”, only one!

• But she didn’t tell me exactly the weather on the days, and how it affected her actions.

• Then I have to figure out by myself! HMM

3

HMM is just a set of 3 rules

• If today weather is then tmrw it will be with probability

• When weather is Jane will do action with probability

• In the 1st day, the weather is with probability https://en.wikipedia.org/wiki/Hidden_Markov_model

4

What are hidden?

• The states of weather {“sunny”, “rainy”} are not observable they are hidden

• The actions {“walk”, “shop”, “clean”} are observed in an index sequence where

5

Two common tasks

1. Given a model and a sequence of action indexes please calculate the probability the model generates the sequence. The forward algorithm

2. Given a sequence , build a model so that is maximum. The Baum-Welch algorithm

6

The forward algorithm

• Let be the probability of generating the sequence and ending up at state

• Using dynamic programming we have:

• And result:

7

The Baum-Welch algorithm

• Given a model , we use it to generate many sequences, but consider only the ones that emit :

Main idea: init with a random model and make it better incrementally

• Nothing is hidden in these sequences! Now we simply base on them to estimate

8

Estimate

• To estimate , count the transitions from to and to other states

• To estimate , count the appearances of that have action index , also count all the appearances of

• To estimate , count the appearances of at the first element of all sequences, and count the number of all sequences too

• But, to count all of things above, we need …

9

Forward and backward variables

• Using the forward algorithm we have • Using the backward algorithm we have the

probability of generating the sequence starting from tmrw, given the state of today. Dynamic programming is used again:

10

Estimate

• Count transitions from to :

• Thus:

11

Estimate

• Count the appearances of state :

• Thus:

12

Estimate

• Count the appearances of at the first element:

• Count the number of all sequences:

• Thus:

13

Thank you!

• Contact: katatunix@gmail.com

top related