DRBMC

Trainable classifier by Discriminative Restricted Boltzmann Machine

    W = DRBMC(A,N,L)
    W = A*DRBMC([],N,L)
    W = A*DRBMC(N,L)

Input
A	Dataset
N	Number of hidden units
L	Regularisation parameter (L2)

Output
W	Discriminative Restricted Boltzmann Machine classifier

Description

The classifier trains a discriminative Restricted Boltzmann Machine (RBM) on dataset A. The discriminative RBM can be viewed as a logistic  regressor with hidden units. The discriminative RBM has N hidden units  (default = 5). It is trained with L2 regularisation using regularisation  parameter L (default = 0).

New objects are classified in the same way as in a logistic regressor,  using a softmax function over the labels.

Reference(s)

H. Larochelle and Y. Bengio. Classification using Discriminative Restricted Boltzmann Machines. Proceedings of the 25th International Conference on Machine Learning (ICML), pages 536?543, 2008.

See also

datasets, mappings, loglc,