7. Hidden Markov Models (Parte 2) (pdf, it, 413 KB, 4/28/10)
7. Hidden Markov Models (Parte 2) (pdf, it, 413 KB, 4/28/10)
7. Hidden Markov Models (Parte 2) (pdf, it, 413 KB, 4/28/10)
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Back to the polynomial toy example<br />
-50<br />
-<strong>10</strong>0<br />
-150<br />
-200<br />
-250<br />
0 5 <strong>10</strong> 15 20<br />
-50<br />
-<strong>10</strong>0<br />
-150<br />
-200<br />
-250<br />
0 3 5 <strong>10</strong> 15 20<br />
2<br />
1.5<br />
1<br />
0.5<br />
0<br />
-0.5<br />
-1<br />
-1 -0.5 0 0.5 1<br />
3<br />
2<br />
1<br />
0<br />
-1<br />
estimate<br />
truth<br />
-2<br />
58<br />
-1 -0.5 0 0.5 1<br />
Some more HMM-oriented solutions<br />
• Application driven model selection:<br />
states have some physical meaning<br />
[Hannaford,Lee IJRR 91]<br />
• Spl<strong>it</strong> and merge approaches: starting<br />
from an inappropriate but simple model,<br />
the correct model is determined by<br />
successively applying a spl<strong>it</strong>ting (or<br />
merging) operation<br />
[Ikeda 93] [Singer,Ostendorf ICASSP 96]<br />
[Takami, Sagayama ICASSP 92]<br />
59<br />
6