| |
L1 Norm Minimization for kernel classifiers.
|
Minimize ||\alpha||_norm describing kernel expansion, under
separability constraints. [currently only for pattern recognition and
norms 0, 1 and 2]
A=ANORM(H) returns an anorm object initialized with hyperparameters H.
Hyperparameters, and their defaults:
norm=1 -- norm to use, i.e. minimize ||\alpha||_norm
ridge=1e-12 -- a ridge on the kernel
child=kernel -- the kernel is stored as a member called "child"
Model:
alpha -- the weights
b0 -- the threshold
Xsv -- the Support Vectors
Methods:
train, test
Example:
[r,a]=train(anorm('norm=1'),gen(toy));
[r,a]=train(anorm('norm=2'),gen(toy));
[r,a]=train(anorm({'norm=0',kernel('rbf',1)}),gen(toy));
Refernce : Use of the Zero-Norm With Linear Models and Kernel Methods
| |
Author : Jason Weston, André Elisseeff, Bernd Schölkopf, Mike Tipping |