| |
Abstract:
We present an unsupervised classification algorithm based on
an ICA mixture model. A mixture model is a model in which the
observed data can be categorized into several mutually exclusive
data classes. In an ICA mixture model, it is assumed that the data
in each class are generated by a linear mixture of independent
sources. The algorithm finds the independent sources and the mixing
matrix for each class and also computes the class membership
probability of for each data point. This approach extends the
Gaussian mixture model so that the clusters can have non-Gaussian
structure. Performance on a standard classification problem, the
Iris flower data set, demonstrates that the new algorithm can
improve classification accurately over standard Gaussian mixture
models. We also show that the algorithm can be applied to blind
source separation in nonstationary environments. The method can
switch automatically between learned mixing matrices in different
environments. Preliminary results on natural scenes and text image
patterns show that the algorithm is able to find classes so that
one class encodes the natural images and the other class
specializes on encoding the text segments.
|
