Building Covnets on mnist dataset

#1

it has been said in the video that , no of units should be more in previous dense layer than the no of outputs we are expecting, whats the reason behind this?

#2

Hi, the number of units in previous layer should be more than the number of output classes as we want to predict a class from higher dimensions as coming from a higher dimension to a lower dimension output class, we can easily make a non-linear boundary or hyperplane which distinguishes classes well enough.

#3

Now suppose if the architecture is 3 layered and has input layer with 5 neurons,a hidden layer with 3 neurons and output with 4 neurons. Then going from input layer to hidden layer you decrease the dimensions and the reason might be you think that there are reduntant and extra features, thats okay. But if you now go from low dimensional hidden layer output(3) to the final high dimensional(4) output layer, then it might add some extra non-useful dimension and thus may contibute to wrong classifications.

#4

I hope I’ve cleared your doubt. I ask you to please rate your experience here
Your feedback is very important. It helps us improve our platform and hence provide you
the learning experience you deserve.

On the off chance, you still have some questions or not find the answers satisfactory, you may reopen
the doubt.