Hi,
Yoram Singer and myself wrote a paper about a family of online
algorithms (which can be combined with kernels) exactly for this purpose :
A New Family of Online Algorithms for Category Ranking
Proceedings of the 25rd Conference on Research and Development in
Information Retrieval (SIGIR), 2002. To appear.
Our experiments indicate the these algorithms performs quite well when
compared to one-vs-rest methods.
You can find a copy of it in
http://www.cs.huji.ac.il/~kobics/publication/sigir02.ps.gz .
For more details please contact me.
Regards, Koby.
============================================
Koby Crammer [log in to unmask]
http://www.cs.huji.ac.il/~kobics
============================================
On Wed, 19 Jun 2002, Martin Schulze wrote:
| Hello,
|
| I have a problem with my project at the moment.
| I want to detect the Facial Expressions in a face by means of FACS Codes.
|
| These codes describe facial movements in the face.
| There are 46 codes which describe these movements.
|
| They can appear alone, or in combinations of them.
|
| I am going to use SVM to classify them.
| So i have 46 classes to consider.
|
| No my problem is.
|
| When these classes can appear in combinations, how can the SVM
| classify them.
| It can only classify 1 class at the time.
|
| Can the SVM be modify to classify a dataset to more than 1 class?
| Is it even theoretically possible?
|
| thanks
|
| /Martin
|
|