2011³â Ãß°è Çмú´ëȸ
Current Result Document : 12 / 12
ÇѱÛÁ¦¸ñ(Korean Title) |
MAP Learning of HMM with Practical Dirichlet Priors |
¿µ¹®Á¦¸ñ(English Title) |
MAP Learning of HMM with Practical Dirichlet Priors |
ÀúÀÚ(Author) |
Bong-Kee Sin
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 38 NO. 2(A) PP. 0355 ~ 0356 (2011. 11) |
Çѱ۳»¿ë (Korean Abstract) |
|
¿µ¹®³»¿ë (English Abstract) |
Given a small or very small training set, it is painfully difficult to train HMM reliably without overfitting. This paper presents a very useful method of MAP learning of multinomial-based discrete HMMs using Dirichlet distributions as conjugate priors. Experimental result shows a very consistent advantage over conventional MLE-based models.
|
Å°¿öµå(Keyword) |
HMM
MAP learning
Dirichlet distribution
Conjuga
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|