The number of neurons in the output layer depends on the number of characters to be recognized. Since this is an attempt to recognize the 26 characters, the number of neurons in the output layer is fixed at 26.
Next: Learning Previous: Back Propagation Algorithm
The recognizing capacity of the neural network lies in the weights
interconnecting the nodes in the different laters. In this network there
is one hidden layer. Since the input consists of an 8*8 matrix file, the total
number of neurons in the input file are 64. In the hiddden layer 128 neurons
are used and that is double the the size of the input. That number was chosen
since the program was not converging with the choice of 20 neurons as it
was getting compressed and losing all it's information. As the time was the
main factor and there were a lot of difficulties in running the program, the
choice was to be on the safe side and hence that number selected. But using so
many neurons in the hidden layer meant using a large database as the number of
weights would increase tremendously and which in turn increses the computation
time. For this network it took a very long time to converge.
The number of neurons in the output layer depends on the number of characters
to be recognized. Since this is an attempt to recognize the 26 characters, the
number of neurons in the output layer is fixed at 26.