* Add the following types of Neural Network:
	
  - Boltzmann machine
  - ART

* Add the ability to perform Principal Component Analysis.   Although
  this is not neural nets per se,  one would normally do it as a
  preprocessing stage.   Since it involves the calculation of
  eigenvectors, it would make sense to use the Octave C++ library (but
  do people want this extra dependency?).

* Extend the multi-layer perceptron to use different training rules
  other than the standard back propogation.

* Support for Radial Basis Functions.



