2018 10-708 (CMU) Probabilistic Graphical Models {Lecture 10} [HHM and CRF]

Posted 2021-02-01 ecoflex

 

 

 

 

 

 

 

 

 

 

 

 

between tags and words, there‘s table 1.

between tags, there‘s table 2.

combine the two tables, p(...) to get the results. 

 

 

 

MRF: factors of the tables not necessarily probabilities

 

BN: must be probabilities. => BN is easier to learn than MRF

 

 

 

 

 

 

 Maximum-Entropy Markov Model (MEMM)

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Marginals:

 

1) forward: 

 

 

 

 

 2) Belief:

 

 

 

HMM is generative, modeling joint probability P(x,y)

but tagging just needs  P(y|x)

 

---

 

https://cedar.buffalo.edu/~srihari/CSE574/Discriminative-Generative.pdf

 

 

 

 

 

---

 

 

Full obervation!

(like the offline SLAM?)

 

 

 

 

 

biased! because we only look at local observation.

 

 

 

 

 

 

 

 

 

 

P（x_2|x_1） can be called Psi(x_1,x_2)

 

 

 

 

If Y_1 ~~~~ Y_{n-2} are connected somehow, what should be changed?

 

 

 

 

 

 

 

 

 

 

How close the model is closed to the truth.