| Who | When |
Messages | |
|
|
|
|
4
|
 |
|
05-17-2008 04:59 AM ET (US)
|
|
Deleted by topic administrator 05-17-2008 10:16 AM
|
| Anup Doshi
|
3
|
|
|
11-03-2005 03:31 PM ET (US)
|
|
My favorite question...why did they choose the DCT as the feature set, and would this formulation work better with a different set of 'words'?
|
| Brendan Morris
|
2
|
|
|
11-03-2005 01:54 PM ET (US)
|
|
I'm fuzzy on the unsupervised learning and what the hidden states L correlate to in this problem. The annotation results are quite impressive. I'm curious what percentage of the dataset had recall = 0 and what might be the reasons for this.
As a side note it's amazing how taking Vasconcelos class helps understand the Bayesian formulation.
|
| Robin Hewitt
|
1
|
|
|
11-03-2005 10:14 AM ET (US)
|
|
How are the authors choosing the value for K in the EM algorithm...is K set to the number of semantic concepts?
|
