Dense layer output dimensions:

#1

As i have asked about my previous doubt, for LeNet the last covnet layer is 400*1 which will act as input to the next dense layer.

BUT WHY THE OUTPUT OF DENSE LAYER CHOOSING IS 400 AND 84 ONLY?
like if we want to apply this for digits recognition, then the DENSE LAYER OUTPUT SHOULD BE 10.

SO I WANT TO KNOW THE CHOOSING CRITERIA FOR DENSE LAYER OUTPUT DIMENSION -----LIKE IT IS FIXED THAT WE HAVE TO USE 400 AND 84 OR IT CAN BE VARIED ACCORDING TO OUR NEED LIKE IN DIGITS RECOGNITION OUR CLASSES ARE 10 SO 10 OUTPUTS???

#2

As I ALREADY EXPLAINED IN YOUR PREVIOUS DOUBT THAT THE OUTPUT LAYER SIZE DEPENDS ON YOUR OUTPUT CLASSES THAT YOU WANT TO CLASSIFY.
IT DEPENDS ON THE USE CASES. FOR MNIST 10 WILL BE OUTPUT LAYER SIZE
BUT YOU ASKED ABOUT LENET SO I EXPLAINED WITH RESPECT TO LENET AND ALEXNET.
FOR MORE INFORMATION REFER THE RESEARCH PAPER AND ARCHITECTURE.