Three basic problems encountered during the training of the network are: (i) The number of
output nodes is not known a priori, (ii) iteration number and how to change the parameter
values during the training are not known, and (iii) too many nodes are used to represent the
distribution.