This is a demo program for the back-propagation algorithm.  The program uses
back-prop to learn the feature-to-letter mapping. 

It uses number_generators (number_generators.f90)

A 14-feature definition for each upper-case letter is provided in ltrdefs (see
Rummelhart & McClelland (1981, Psych. Rev).


Requires: number_generators (number_generators.f90)

Output: a confusion matrix for each epoch with a count of the errors.  Sample
output is provided in FA.out


Tutorial for back-prop on the web:
	
     	www.speech.sri.com/people/anand/771/html/node37.html

Files:
	FA.f90:  code
	
	number_generators.f90:	F90 module with random number generators 
		and filters for Gaussian, exponential and other functions
		
	ltrdefs:  features for each letter; -1 not present, +1 = present
	
	FA.out:  sample output